{{Short description|Autonomous artificial intelligence agent}} {{Unreliable sources|date=December 2025}} {{Use mdy dates|date=April 2025}}
In the context of generative artificial intelligence, '''AI agents''' (also referred to as '''compound AI systems''' or '''agentic AI''') are a class of intelligent agents that can pursue goals, use tools, and take actions with varying degrees of autonomy. In practice, they usually operate within human-defined objectives, constraints, and available tools.<ref name="NIST-AI-agent-standards">{{Cite web |title=AI Agent Standards Initiative |url=https://www.nist.gov/artificial-intelligence/ai-agent-standards-initiative |website=NIST |access-date=2026-05-05}}</ref><ref name="IBM-AI-agents">{{Cite web |title=What Are AI Agents? |url=https://www.ibm.com/think/topics/ai-agents |website=IBM |access-date=2026-05-05}}</ref>
==Overview== AI agents possess several key attributes, including goal-directed behavior, natural language interfaces, the capacity to use external tools, and the ability to perform multi-step tasks. Their control flow is frequently driven by large language models (LLMs). Agent systems may also include memory components, planning logic, tool interfaces, and orchestration software for coordinating agent components.<ref name="IBM-AI-agents" /><ref name="GoogleCloud-AI-agents">{{Cite web |title=What are AI agents? Definition, examples, and types |url=https://cloud.google.com/discover/what-are-ai-agents |website=Google Cloud |access-date=2026-05-05}}</ref>
AI agents do not have a standard definition.<ref>{{Cite web |last1=Zeff|first1=Maxwell|last2=Wiggers|first2=Kyle|date=2025-03-14|title=No one knows what the hell an AI agent is|url=https://techcrunch.com/2025/03/14/no-one-knows-what-the-hell-an-ai-agent-is/|archive-url=https://web.archive.org/web/20250318134231/https://techcrunch.com/2025/03/14/no-one-knows-what-the-hell-an-ai-agent-is/|archive-date=2025-03-18|access-date=2025-05-15|website=TechCrunch|language=en-US}}</ref><ref>{{Cite web |last=Varanasi|first=Lakshmi|title=AI agents are all the rage. But no one can agree on what they do.|url=https://www.businessinsider.com/what-is-an-ai-agent-depends-who-you-ask-2025-3|archive-url=https://web.archive.org/web/20250411143511/https://www.businessinsider.com/what-is-an-ai-agent-depends-who-you-ask-2025-3|archive-date=2025-04-11|access-date=2025-05-15|website=Business Insider|language=en-US}}</ref><ref>{{Cite web |last=Bort|first=Julie|date=2025-05-12|title=Even a16z VCs say no one really knows what an AI agent is|url=https://techcrunch.com/2025/05/12/even-a16z-vcs-say-no-one-really-knows-what-an-ai-agent-is/|archive-url=https://web.archive.org/web/20250512184704/https://techcrunch.com/2025/05/12/even-a16z-vcs-say-no-one-really-knows-what-an-ai-agent-is/|archive-date=2025-05-12|access-date=2025-05-15|website=TechCrunch|language=en-US}}</ref> NIST describes agentic AI as an emerging area requiring standards for secure operation, interoperability, and reliable interaction with external systems.<ref name="NIST-AI-agent-standards" />
A common application of AI agents is task automation: for example, booking travel plans based on a user's prompted request.<ref name="The National Law Review-2024">{{Cite web |date=2024-12-30|title=AI Agents: The Next Generation of Artificial Intelligence|url=https://natlawreview.com/article/next-generation-ai-here-come-agents|archive-url=https://web.archive.org/web/20250111192703/https://natlawreview.com/article/next-generation-ai-here-come-agents|archive-date=2025-01-11|access-date=2025-01-14|website=The National Law Review|language=en}}</ref><ref name="World Economic Forum-2024">{{Cite web |date=2024-12-16|title=What are the risks and benefits of 'AI agents'?|url=https://www.weforum.org/stories/2024/12/ai-agents-risks-artificial-intelligence/|archive-url=https://web.archive.org/web/20241228013835/https://www.weforum.org/stories/2024/12/ai-agents-risks-artificial-intelligence/|archive-date=2024-12-28|access-date=2025-01-14|website=World Economic Forum|language=en}}</ref><ref name="Knight-2024">{{Cite magazine |last=Knight|first=Will|date=2024-03-14|title=Forget Chatbots. AI Agents Are the Future|url=https://www.wired.com/story/fast-forward-forget-chatbots-ai-agents-are-the-future/|archive-url=https://web.archive.org/web/20250105095231/https://www.wired.com/story/fast-forward-forget-chatbots-ai-agents-are-the-future/|archive-date=2025-01-05|access-date=2025-01-14|magazine=Wired|language=en-US|issn=1059-1028}}</ref>
Companies such as Google, Microsoft and Amazon Web Services have offered platforms for deploying pre-built AI agents.<ref>{{Cite web |last=David|first=Emilia|date=2025-04-15|title=Moveworks joins AI agent library craze|url=https://venturebeat.com/ai/moveworks-joins-ai-agent-library-craze/|archive-url=https://web.archive.org/web/20250415214729/https://venturebeat.com/ai/moveworks-joins-ai-agent-library-craze/|archive-date=2025-04-15|access-date=2025-05-14|website=VentureBeat|language=en-US}}</ref> Several protocols have been proposed for standardizing inter-agent communication, with examples including the Model Context Protocol, Gibberlink,<ref>{{Cite web |last=Zeff|first=Maxwell|date=2025-03-05|title=GibberLink lets AI agents call each other in robo-language|url=https://techcrunch.com/2025/03/05/gibberlink-lets-ai-agents-call-each-other-in-robo-language/|archive-url=https://web.archive.org/web/20250305141006/https://techcrunch.com/2025/03/05/gibberlink-lets-ai-agents-call-each-other-in-robo-language/|archive-date=2025-03-05|access-date=2025-04-02|website=TechCrunch|language=en-US}}</ref> and many others. Some of these protocols are also used for connecting agents to external applications.<ref name="Holmes-2025a">{{Cite web |last=Holmes |first=Aaron |date=2025-07-07 |title=The Seven Kinds of AI Agents |url=https://www.theinformation.com/articles/seven-kinds-ai-agents |archive-url=https://archive.today/20250720144318/https://www.theinformation.com/articles/seven-kinds-ai-agents |archive-date=2025-07-20 |access-date=2025-11-09 |website=The Information |language=en}}</ref>
In December 2025, Linux Foundation announced the formation of the Agentic AI Foundation (AAIF), with the goal of ensuring agentic AI evolves transparently and collaboratively.<ref>{{cite press release |title=Linux Foundation Announces the Formation of the Agentic AI Foundation (AAIF), Anchored by New Project Contributions Including Model Context Protocol (MCP), goose and AGENTS.md |url=https://aaif.io/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation-aaif-anchored-by-new-project-contributions-including-model-context-protocol-mcp-goose-and-agents-md/ |website=Agentic AI Foundation |access-date=2026-01-17}}</ref><ref>{{Cite news |title=Linux Foundation founds Agentic AI Foundation |url=https://www.theregister.com/2025/12/09/linux_foundation_agentic_ai_foundation/ |archive-url=http://web.archive.org/web/20251228193316/https://www.theregister.com/2025/12/09/linux_foundation_agentic_ai_foundation/ |archive-date=2025-12-28 |access-date=2026-01-22 |language=en}}</ref>
==History== {{Main|History of artificial intelligence}}
AI agents have been traced back to research from the 1990s, with Harvard professor Milind Tambe noting that the definition of an AI agent was not clear at the time. Researcher Andrew Ng has been credited with spreading the term "agentic" to a wider audience in 2024.<ref>{{Cite web|title=What does 'agentic' AI mean? Tech's newest buzzword is a mix of marketing fluff and real promise|url=https://apnews.com/article/agentic-ai-agents-microsoft-amazon-518d6ae159d1f4d3343e98a456cb5221|website=Associated Press|date=2025-11-18|access-date=2025-11-28|language=en|last=O'Brien|first=Matt|archive-url=https://web.archive.org/web/20251118171507/https://apnews.com/article/agentic-ai-agents-microsoft-amazon-518d6ae159d1f4d3343e98a456cb5221|archive-date=2025-11-18}}</ref>
== Training and testing == Researchers have attempted to build world models<ref>{{Cite magazine |last=Knight|first=Will|date=2025-05-22|title=A United Arab Emirates Lab Announces Frontier AI Projects—and a New Outpost in Silicon Valley|url=https://www.wired.com/story/the-united-arab-emirates-announces-frontier-ai-projects-and-a-new-lab-in-silicon-valley/|archive-url=https://web.archive.org/web/20250522202354/https://www.wired.com/story/the-united-arab-emirates-announces-frontier-ai-projects-and-a-new-lab-in-silicon-valley/|archive-date=2025-05-22|access-date=2025-11-09|magazine=Wired|language=en-US|issn=1059-1028}}</ref><ref>{{Cite web |last=Orland|first=Kyle|date=2024-12-06|title=Google's Genie 2 "world model" reveal leaves more questions than answers|url=https://arstechnica.com/ai/2024/12/googles-genie-2-world-model-reveal-leaves-more-questions-than-answers/|archive-url=https://web.archive.org/web/20241207000413/https://arstechnica.com/ai/2024/12/googles-genie-2-world-model-reveal-leaves-more-questions-than-answers/|archive-date=2024-12-07|access-date=2025-11-09|website=Ars Technica|language=en}}</ref> and reinforcement learning environments<ref>{{Cite web |last=Zeff|first=Maxwell|date=2025-09-21|title=Silicon Valley bets big on 'environments' to train AI agents|url=https://techcrunch.com/2025/09/21/silicon-valley-bets-big-on-environments-to-train-ai-agents/|archive-url=https://web.archive.org/web/20250916191353/https://techcrunch.com/2025/09/16/silicon-valley-bets-big-on-environments-to-train-ai-agents/|archive-date=2025-09-16|access-date=2025-11-09|website=TechCrunch|language=en-US}}</ref> to train or evaluate AI agents. For example, video games such as Minecraft<ref>{{Cite web |last=Shazhaev |first=Ilman |date=2025-11-24 |title=Why Game Engines Are Becoming A.I.'s Most Important Testbeds |url=https://observer.com/2025/11/gaming-training-next-gen-ai/ |archive-url=https://web.archive.org/web/20251203023639/https://observer.com/2025/11/gaming-training-next-gen-ai/ |archive-date=2025-12-03 |access-date=2025-12-03 |website=Observer |language=en-US}}</ref> and No Man's Sky<ref>{{Cite web |last=David |first=Emilia |date=2024-03-13 |title=Google's new AI will play video games with you — but not to win |url=https://www.theverge.com/2024/3/13/24099024/google-deepmind-ai-agent-sima-video-games |archive-url=https://web.archive.org/web/20250602020104/https://www.theverge.com/2024/3/13/24099024/google-deepmind-ai-agent-sima-video-games |archive-date=2025-06-02 |access-date=2025-12-03 |website=The Verge |language=en-US}}</ref> as well as replicas of company websites,<ref>{{Cite web |last=Metz |first=Cade |date=2025-12-02 |title=Silicon Valley Builds Amazon and Gmail Copycats to Train A.I. Agents |url=https://www.nytimes.com/2025/12/02/technology/artificial-intelligence-amazon-gmail.html |archive-url=https://archive.today/20251202165659/https://www.nytimes.com/2025/12/02/technology/artificial-intelligence-amazon-gmail.html |archive-date=2025-12-02 |access-date=2025-12-02 |website=The New York Times}}</ref> have also been used for training such agents.
== Autonomous capabilities == The ''Financial Times'' compared the autonomy of AI agents to the SAE classification of self-driving cars, likening most applications to level 2 or level 3, with some achieving level 4 in highly specialized circumstances, and level 5 being theoretical.<ref>{{Cite news |last=Colback|first=Lucy|date=2025-05-07|title=AI agents: from co-pilot to autopilot|url=https://www.ft.com/content/3e862e23-6e2c-4670-a68c-e204379fe01f|archive-url=https://archive.today/20250507031905/https://www.ft.com/content/3e862e23-6e2c-4670-a68c-e204379fe01f|archive-date=2025-05-07|access-date=2025-05-14|work=Financial Times}}</ref>
== Cognitive architecture == {{See also|Large language model#Agency}} The following are some internal design options for reasoning within an agent:<ref>{{Cite journal |last1=Liu |first1=Yue |last2=Lo |first2=Sin Kit |last3=Lu |first3=Qinghua |last4=Zhu |first4=Liming |last5=Zhao |first5=Dehai |last6=Xu |first6=Xiwei |last7=Harrer |first7=Stefan |last8=Whittle |first8=Jon |date=2025-02-01 |title=Agent design pattern catalogue: A collection of architectural patterns for foundation model based agents |journal=Journal of Systems and Software |volume=220 |article-number=112278 |doi=10.1016/j.jss.2024.112278 |issn=0164-1212|doi-access=free }}</ref> * Retrieval-augmented generation * ReAct (Reason + Act) pattern is an iterative process in which an AI agent alternates between reasoning and taking actions, receives observations from the environment or external tools, and integrates these observations into subsequent reasoning steps.<ref>{{cite conference |last1=Yao |first1=Shuai |last2=Zhao |first2=Jie |last3=Yu |first3=Dong |last4=Du |first4=Nicholas |last5=Shafran |first5=Itamar |last6=Narasimhan |first6=Karthik |last7=Cao |first7=Yizhe |year=2023 |title=REACT: SYNERGIZING REASONING AND ACTING IN LANGUAGE MODELS |url=https://openreview.net/forum?id=WE_vluYUL-X |conference=11th International Conference on Learning Representations (ICLR 2023) |location=Kigali, Rwanda}}</ref><ref>{{cite arXiv |last1=Yao |first1=Shunyu |last2=Zhao |first2=Jeffrey |last3=Yu |first3=Dian |last4=Du |first4=Nan |last5=Shafran |first5=Izhak |last6=Narasimhan |first6=Karthik |last7=Cao |first7=Yuan |title=ReAct: Synergizing Reasoning and Acting in Language Models |eprint=2210.03629 |class=cs.CL |year=2022}}</ref> * Reflexion, which uses an LLM to create feedback on the agent's plan of action and stores that feedback in a memory cache. * A tool/agent registry, for organizing software functions or other agents that the agent can use. * One-shot model querying, which queries the model once to create the plan of action.
=== Reference architecture === Ken Huang proposed an AI agent reference architecture, which consists of seven interconnected layers, with each layer building on the functionality of the layers beneath it<ref>{{cite book |last1=Huang |first1=Ken |title=Agentic AI: theories and practices |date=2025 |publisher=Springer |location=Cham |isbn=978-3-031-90025-9}}</ref>: * '''Layer 1''': Foundation models - provide the core AI engines to power agent capabilities. * '''Layer 2''': Data operations - manage the complex data infrastructure required for AI agent operations, including Vector database, data loaders, RAG. * '''Layer 3''': Agent frameworks - sophisticated software and tools that simplify the development and management of the AI agents. * '''Layer 4''': Deployment and infrastructure - provide the robust technical foundation for running AI agents. * '''Layer 5''': Evaluation and observability - focus on assessing the safety and performance of AI agents. * '''Layer 6''': Security and compliance - a crucial protective framework ensuring AI agents operate safely, securely, and conform to regulatory boundaries. At this layer security and compliance features embedded into all the AI agent stack layers are integrated together. * '''Layer 7''': Agent ecosystem - represents the AI agents' interface with real-world applications and users.
== Orchestration patterns == To execute complex tasks, autonomous agents are often integrated with other agents or specialized tools. These configurations, known as orchestration patterns or workflows, include the following:<ref>{{Cite book |last=Gullí |first=Antonio |title=Agentic Design Patterns A Hands-On Guide to Building Intelligent Systems |date=October 30, 2025 |publisher=Springer |isbn=9783032014023}}</ref><ref>{{Cite book |last=Kar |first=Indrajit |title=Building Multimodal Generative AI and Agentic Applications Shaping Concept to Code for the Future of Multimodal and Advanced Agentic GenAI Applications |date=October 27, 2025 |publisher=BPB Publications |isbn=9789365898385}}</ref> * '''Prompt chaining''': A sequence where the output of one step serves as the input for the next. * '''Routing''': The classification of an input to direct it to a specialized downstream task or tool. * '''Parallelization''': The simultaneous execution of multiple tasks. * '''Sequential processing''': A fixed, linear progression of tasks through a predefined pipeline. * '''Planner-critic''': An iterative pattern where one agent generates a proposal and another evaluates it to provide feedback for refinement.
== Multimodal AI agents == In addition to large language models (LLMs), vision-language models (VLMs) and multimodal foundation models can be used as the basis for agents. In September 2024, Allen Institute for AI released an open-source vision-language model.<ref>{{Cite magazine |last=Knight|first=Will|date=2024-09-25|title=The Most Capable Open Source AI Model Yet Could Supercharge AI Agents|url=https://www.wired.com/story/molmo-open-source-multimodal-ai-model-allen-institute-agents/|archive-url=https://web.archive.org/web/20250328102342/https://www.wired.com/story/molmo-open-source-multimodal-ai-model-allen-institute-agents/|archive-date=2025-03-28|access-date=2025-06-12|magazine=Wired|language=en-US|issn=1059-1028}}</ref> Nvidia released a framework for developers to use VLMs, LLMs and retrieval-augmented generation for building AI agents that can analyze images and videos, including video search and video summarization.<ref>{{Cite web |last=Takahashi|first=Dean|date=2024-11-04|title=Nvidia AI Blueprint makes it easy for any devs to build automated agents that analyze video|url=https://venturebeat.com/ai/nvidia-ai-blueprint-makes-it-easy-for-devs-in-any-industry-build-agents-to-analyze-video/|archive-url=https://web.archive.org/web/20241205103955/https://venturebeat.com/ai/nvidia-ai-blueprint-makes-it-easy-for-devs-in-any-industry-build-agents-to-analyze-video/|archive-date=2024-12-05|access-date=2025-06-12|website=VentureBeat|language=en-US}}</ref><ref>{{Cite web |last=Takahashi|first=Dean|date=2025-01-07|title=Nvidia launches blueprint for AI agents that can analyze video|url=https://venturebeat.com/ai/nvidia-launches-blueprint-for-ai-agents-that-can-analyze-video/|archive-url=https://web.archive.org/web/20250404224324/https://venturebeat.com/ai/nvidia-launches-blueprint-for-ai-agents-that-can-analyze-video/|archive-date=2025-04-04|access-date=2025-06-12|website=VentureBeat|language=en-US}}</ref> Microsoft released a multimodal agent model – trained on images, video, software user interface interactions, and robotics data – that the company claimed can manipulate software and robots.<ref>{{Cite web |last=Edwards|first=Benj|date=2025-02-20|title=Microsoft's new AI agent can control software and robots|url=https://arstechnica.com/ai/2025/02/microsofts-new-ai-agent-can-control-software-and-robots/|archive-url=https://web.archive.org/web/20250520062232/https://arstechnica.com/ai/2025/02/microsofts-new-ai-agent-can-control-software-and-robots/|archive-date=2025-05-20|access-date=2025-06-12|website=Ars Technica|language=en}}</ref>
==Applications== {{See also|Lists of open-source artificial intelligence software#Agentic AI|l1=List of open-source agentic AI handlers}} As of April 2025, per the Associated Press, there are few real-world applications of AI agents.<ref>{{Cite web |date=2025-04-30|title=Visa wants to give artificial intelligence 'agents' your credit card|url=https://apnews.com/article/ai-artificial-intelligence-5dfa1da145689e7951a181e2253ab349|archive-url=https://web.archive.org/web/20250501010808/https://apnews.com/article/ai-artificial-intelligence-5dfa1da145689e7951a181e2253ab349|archive-date=2025-05-01|access-date=2025-05-14|website=Associated Press|language=en}}</ref> As of June 2025, per ''Fortune'', many companies are primarily experimenting with AI agents.<ref name="Goldman-2025">{{Cite web |last=Goldman|first=Sharon|date=2025-06-11|title=Microsoft Copilot flaw raises urgent questions for any business deploying AI agents|url=https://fortune.com/2025/06/11/microsoft-copilot-vulnerability-ai-agents-echoleak-hacking/|archive-url=https://web.archive.org/web/20250611235202/https://fortune.com/2025/06/11/microsoft-copilot-vulnerability-ai-agents-echoleak-hacking/|archive-date=2025-06-11|access-date=2025-06-12|website=Fortune|language=en}}</ref>
''The Information'' divided AI agents into seven archetypes: business-task agents, for acting within enterprise software; conversational agents, which act as chatbots for customer support; research agents, for querying and analyzing information (such as OpenAI Deep Research); analytics agents, for analyzing data to create reports; software developer or coding agents (such as Cursor); domain-specific agents, which include specific subject matter knowledge; and web browser agents (such as OpenAI Operator).<ref name="Holmes-2025a" />
By mid-2025, AI agents have been used in video game development,<ref>{{Cite web |last=Kachwala|first=Zaheer|date=2025-08-18|title=Nearly 90% of videogame developers use AI agents, Google study shows|url=https://www.reuters.com/business/nearly-90-videogame-developers-use-ai-agents-google-study-shows-2025-08-18/|archive-url=https://archive.today/20250818153431/https://www.reuters.com/business/nearly-90-videogame-developers-use-ai-agents-google-study-shows-2025-08-18/|archive-date=2025-08-18|access-date=2025-11-09|website=Reuters}}</ref> gambling (including sports betting),<ref name="Knibbs-2025">{{Cite magazine |last=Knibbs|first=Kate|date=2025-09-02|title=Meet the Guys Betting Big on AI Gambling Agents|url=https://www.wired.com/story/sports-betting-crypto-artificial-intelligence-agents/|archive-url=https://web.archive.org/web/20250902112628/https://www.wired.com/story/sports-betting-crypto-artificial-intelligence-agents/|archive-date=2025-09-02|access-date=2025-11-09|magazine=Wired|language=en-US|issn=1059-1028}}</ref> cryptocurrency wallets<ref name="Knibbs-2025" /> (including cryptocurrency trading and meme coins<ref name="Kharif-2025">{{Cite web |last=Kharif|first=Olga|date=2025-07-29|title=Cornell Tech Professor Warns AI Agents And Crypto Spell Trouble|url=https://www.bloomberg.com/news/newsletters/2025-07-29/cornell-tech-professor-says-ai-agents-and-crypto-spell-trouble-bloomberg-crypto|archive-url=https://archive.today/20250729233851/https://www.bloomberg.com/news/newsletters/2025-07-29/cornell-tech-professor-says-ai-agents-and-crypto-spell-trouble-bloomberg-crypto|archive-date=2025-07-29|access-date=2025-11-09|website=Bloomberg News}}</ref>) and social media.<ref>{{Cite web |last=Scarcella |first=Mike |date=2025-12-30 |title=Musk's X resolves Eliza Labs lawsuit over AI agents |url=https://www.reuters.com/legal/government/musks-x-resolves-eliza-labs-lawsuit-over-ai-agents-2025-12-30/ |archive-url=https://archive.today/20251230160412/https://www.reuters.com/legal/government/musks-x-resolves-eliza-labs-lawsuit-over-ai-agents-2025-12-30/ |archive-date=2025-12-30 |access-date=2026-01-07 |website=Reuters}}</ref> In August 2025, ''New York Magazine'' described software development as the most definitive use case of AI agents.<ref name="Herrman-2025">{{Cite web |last=Herrman|first=John|date=2025-08-22|title=Why Everything's an AI 'Agent' Now|url=https://nymag.com/intelligencer/article/why-everything-is-an-ai-agent-now.html|archive-url=https://web.archive.org/web/20250822141745/https://nymag.com/intelligencer/article/why-everything-is-an-ai-agent-now.html|archive-date=2025-08-22|access-date=2025-11-09|website=New York|language=en}}</ref> Likewise, by October 2025, noting a decline in expectations, ''The Information'' noted AI coding agents and customer support as the primary use cases by businesses.<ref name="Holmes-2025b">{{Cite web |last=Holmes|first=Aaron|date=2025-10-21|title=A Reality Check on Agents|url=https://www.theinformation.com/articles/reality-check-agents|archive-url=https://archive.today/20251022154702/https://www.theinformation.com/articles/reality-check-agents|archive-date=2025-10-22|access-date=2025-11-09|website=The Information|language=en}}</ref>
In November 2025, ''The Wall Street Journal'' reported that few companies that deployed AI agents have received a return on investment.<ref>{{Cite news|title=Companies Begin to See a Return on AI Agents|url=https://www.wsj.com/articles/companies-begin-to-see-a-return-on-ai-agents-671d830d|work=The Wall Street Journal|date=2025-11-12|access-date=2025-11-28|issn=0099-9660|language=en-US|first=Steven|last=Rosenbush|archive-url=https://archive.today/20251124000338/https://www.wsj.com/articles/companies-begin-to-see-a-return-on-ai-agents-671d830d|archive-date=November 24, 2025}}</ref>
=== Applications in government === Several government bodies in the United States and United Kingdom have deployed or announced the deployment of agents, at the local and national level. The city of Kyle, Texas deployed an AI agent from Salesforce in March 2025 for 311 customer service.<ref>{{Cite web |last=Alms |first=Natalie |date=2025-10-23 |title=As agencies shed staff, industry execs predict AI agents' rise |url=https://www.govexec.com/technology/2025/10/salesforce-pitches-ai-agents-government-sheds-staff/409017/ |archive-url=https://archive.today/20251207192203/https://www.govexec.com/technology/2025/10/salesforce-pitches-ai-agents-government-sheds-staff/409017/ |archive-date=2025-12-07 |access-date=2025-12-11 |website=Government Executive |language=en}}</ref> In November 2025, the Internal Revenue Service stated that it would use Agentforce, AI agents from Salesforce, for the Office of Chief Counsel, Taxpayer Advocate Services and the Office of Appeals.<ref>{{Cite web|title=Exclusive: IRS deploys AI agents|url=https://www.axios.com/2025/11/21/irs-deploys-ai-agents|website=Axios|date=2025-11-21|access-date=2025-11-28|language=en|first=Ashley|last=Gold|archive-url=https://archive.today/20251121125008/https://www.axios.com/2025/11/21/irs-deploys-ai-agents|archive-date=2025-11-21}}</ref> That same month, Staffordshire Police announced that they would trial Agentforce agents for handling non-emergency 101 calls in the United Kingdom starting in 2026.<ref>{{Cite web|title=Staffordshire Police to trial AI 'agents' on 101 service|url=https://www.bbc.com/news/articles/cvgq03y0l3yo|website=BBC|date=2025-11-26|access-date=2025-11-28|language=en-GB|last=Corrigan|first=Phil|archive-url=https://archive.today/20251127051703/https://www.bbc.com/news/articles/cvgq03y0l3yo|archive-date=2025-11-27}}</ref> In December 2025, the Department of Neighborhoods in Detroit, Michigan, in partnership with a local business, deployed a pilot project in two Detroit districts for an AI agent to be used for customer service calls.<ref>{{Cite web |last=Fernandez |first=Demond |date=2025-12-12 |title=Residents in 2 Detroit districts now assisted by AI agent for city service calls |url=https://www.clickondetroit.com/news/local/2025/12/12/residents-in-2-detroit-districts-now-assisted-by-ai-agent-for-city-service-calls/ |archive-url=https://web.archive.org/web/20251217113711/https://www.clickondetroit.com/news/local/2025/12/12/residents-in-2-detroit-districts-now-assisted-by-ai-agent-for-city-service-calls/ |archive-date=2025-12-17 |access-date=2025-12-17 |website=WDIV |language=en}}</ref>
In February 2025, Thomas Shedd, the director of the Technology Transformation Services, proposed using AI coding agents across the United States federal government.<ref>{{Cite web |last=Wong |first=Matteo |date=2025-03-10 |title=DOGE's Plans to Replace Humans With AI Are Already Under Way |url=https://www.theatlantic.com/technology/archive/2025/03/gsa-chat-doge-ai/681987/ |archive-url=https://web.archive.org/web/20251204163102/https://www.theatlantic.com/technology/archive/2025/03/gsa-chat-doge-ai/681987/ |archive-date=2025-12-04 |access-date=2025-12-25 |website=The Atlantic |language=en}}</ref> A recruiter for the Department of Government Efficiency proposed in April 2025 to use AI agents to automate the work of about 70,000 United States federal government employees, as part of a startup with funding from OpenAI and a partnership agreement with Palantir. This proposal was criticized by experts for its impracticality, if not impossibility, and the lack of corresponding widespread adoption by businesses.<ref>{{Cite magazine |last=Haskins |first=Caroline |date=2025-05-02 |title=A DOGE Recruiter Is Staffing a Project to Deploy AI Agents Across the US Government |url=https://www.wired.com/story/doge-recruiter-ai-agents-palantir-clown-emoji/ |archive-url=https://web.archive.org/web/20250503074840/https://www.wired.com/story/doge-recruiter-ai-agents-palantir-clown-emoji/ |archive-date=2025-05-03 |access-date=2025-05-14 |magazine=Wired |language=en-US |issn=1059-1028}}</ref>
In December 2025, the Food and Drug Administration announced that it would offer "agentic AI capabilities" to its staff for "meeting management, pre-market reviews, review validation, post-market surveillance, inspections and compliance and administrative functions."<ref>{{Cite web |last=Aguilar |first=Mario |date=2025-12-01 |title=FDA offers staff 'agentic AI' to support premarket reviews, administrative tasks |url=https://www.statnews.com/2025/12/01/fda-announces-agentic-ai-elsa-pre-merket-review/ |archive-url=https://web.archive.org/web/20251202002221/https://www.statnews.com/2025/12/01/fda-announces-agentic-ai-elsa-pre-merket-review/ |archive-date=2025-12-02 |access-date=2025-12-06 |website=STAT |language=en-US}}</ref> That same month, the United States Department of Defense launched GenAI.mil, an internal platform for American military personnel to use generative AI-based applications based on Google Gemini, including "intelligent agentic workflows". Defense Secretary Pete Hegseth listed applications such as "[conducting] deep research, [formatting] documents and even [analyzing] video or imagery at unprecedented speed."<ref>{{Cite web |last=Losey |first=Stephen |date=2025-12-09 |title=Pentagon taps Google Gemini, launches new site to boost AI use |url=https://www.defensenews.com/pentagon/2025/12/09/pentagon-taps-google-gemini-launches-new-site-to-boost-ai-use/ |archive-url=https://archive.today/20251210024513/https://www.defensenews.com/pentagon/2025/12/09/pentagon-taps-google-gemini-launches-new-site-to-boost-ai-use/ |archive-date=2025-12-10 |access-date=2025-12-11 |website=Defense News |language=en}}</ref> In December 2025, the United States Immigration and Customs Enforcement agency signed a contract with a company for its Enforcement and Removal Operations department to use AI agents for skip tracing.<ref>{{Cite web |last=Cox |first=Joseph |date=2025-12-18 |title=ICE Contracts Company Making Bounty Hunter AI Agents |url=https://www.404media.co/ice-contracts-company-making-bounty-hunter-ai-agents/ |archive-url=https://archive.today/20251218173355/https://www.404media.co/ice-contracts-company-making-bounty-hunter-ai-agents/ |archive-date=2025-12-18 |access-date=2025-12-21 |website=404 Media |language=en}}</ref>
=== Operating systems === AI agents have also been integrated into operating systems. Agents have been included in operating systems developed by Microsoft, Apple and Google.<ref name="The Economist-2025" /> In November 2025, Microsoft released a test software build of Windows 11 that included agents intended to run background tasks, with the ability to read and write personal files.<ref>{{Cite web |last=Cunningham |first=Andrew |date=2025-11-18 |title=Microsoft tries to head off the "novel security risks" of Windows 11 AI agents |url=https://arstechnica.com/gadgets/2025/11/new-windows-11-ai-agents-can-work-in-the-background-but-create-new-security-risks/ |archive-url=https://archive.today/20251119121116/https://arstechnica.com/gadgets/2025/11/new-windows-11-ai-agents-can-work-in-the-background-but-create-new-security-risks/ |archive-date=2025-11-19 |access-date=2025-11-28 |website=Ars Technica |language=en}}</ref> In December 2025, ByteDance released Doubao, an AI agent that can be integrated into smartphone operating systems, particularly the Nubia M153 by ZTE.<ref name="Yang-2025">{{Cite magazine |last=Yang |first=Zeyi |date=2025-12-04 |title=ByteDance and DeepSeek Are Placing Very Different AI Bets |url=https://www.wired.com/story/deepseek-goes-high-while-bytedance-goes-wide/ |archive-url=https://web.archive.org/web/20251205023422/https://www.wired.com/story/deepseek-goes-high-while-bytedance-goes-wide/ |archive-date=2025-12-05 |access-date=2025-12-11 |magazine=Wired |language=en-US |issn=1059-1028}}</ref> Several apps in China blocked or restricted the agent, citing privacy and security concerns,<ref>{{Cite web |last=Chourasia |first=Ayush |date=2025-12-08 |title=What is the TikTok owner's Agent AI phone? Find out why is it facing backlash in China |url=https://me.mashable.com/tech/64373/what-is-the-tiktok-owners-agent-ai-phone-find-out-why-is-it-facing-backlash-in-china |archive-url=https://web.archive.org/web/20251210032356/https://me.mashable.com/tech/64373/what-is-the-tiktok-owners-agent-ai-phone-find-out-why-is-it-facing-backlash-in-china |archive-date=2025-12-10 |access-date=2025-12-11 |website=Mashable |language=en-ae}}</ref> including WeChat,<ref name="Yang-2025" /> Alipay, Taobao, Pinduoduo, Ele.me,<ref>{{Cite web |last=Xu |first=Eunice |date=2025-12-07 |title=ByteDance's agentic AI smartphone dials up a digital backlash by China's top apps |url=https://www.scmp.com/business/china-business/article/3335404/bytedances-agentic-ai-smartphone-dials-digital-backlash-chinas-top-apps |archive-url=https://archive.today/20251207024116/https://www.scmp.com/business/china-business/article/3335404/bytedances-agentic-ai-smartphone-dials-digital-backlash-chinas-top-apps |archive-date=2025-12-07 |access-date=2025-12-11 |website=South China Morning Post |language=en}}</ref> and local banks.<ref>{{Cite web |last=Jiang |first=Ben |date=2025-12-09 |title=Z.ai open sources AI agent tool for phones after ByteDance privacy backlash |url=https://www.scmp.com/tech/tech-trends/article/3335746/chinas-zai-open-sources-ai-agent-tool-phones-after-bytedance-privacy-backlash |archive-url=https://archive.today/20251209091231/https://www.scmp.com/tech/tech-trends/article/3335746/chinas-zai-open-sources-ai-agent-tool-phones-after-bytedance-privacy-backlash |archive-date=2025-12-09 |access-date=2025-12-11 |website=South China Morning Post |language=en}}</ref>
=== Web browsing === Web browsers with integrated AI agents are sometimes called agentic browsers. Such agents can perform small tedious tasks during web browsing and potentially even perform browser actions on behalf of the user. Products like OpenAI Operator and Perplexity Comet integrate a spectrum of AI capabilities including the ability to browse the web, interact with websites and perform actions on behalf of the user.<ref>{{Cite web|title=Perplexity launches Comet, an AI-powered web browser|url=https://techcrunch.com/2025/07/09/perplexity-launches-comet-an-ai-powered-web-browser/|website=TechCrunch|date=2025-07-09|access-date=2025-12-01|language=en-US|first=Maxwell|last=Zeff|archive-url=https://web.archive.org/web/20251010111103/https://techcrunch.com/2025/07/09/perplexity-launches-comet-an-ai-powered-web-browser/|archive-date=2025-10-10}}</ref><ref>{{Cite news |last=Roose |first=Kevin |date=February 1, 2025 |title=How Helpful Is Operator, OpenAI's New A.I. Agent? |url=https://www.nytimes.com/2025/02/01/technology/openai-operator-agent.html |access-date=August 9, 2025 |work=The New York Times |language=en}}</ref>
In 2025, Microsoft launched NLWeb, an agentic web search replacement that would allow websites to use agents to query content from websites by using RSS-like interfaces that allow for the lookup and semantic retrieval of content.<ref name="Warren-2025">{{Cite web |last=Warren |first=Tom |date=2025-08-06 |title=Microsoft's plan to fix the web with AI has already hit an embarrassing security flaw |url=https://www.theverge.com/news/719617/microsoft-nlweb-security-flaw-agentic-web |access-date=2025-08-09 |website=The Verge |language=en-US}}</ref> Products integrating agentic web capabilities have been criticised for exfiltrating information about their users to third-party servers and exposing security issues since the way the agents communicate often occur through non-standard protocols.<ref name="Warren-2025" />
== Proposed benefits == AI agents have been proposed as a means of increasing personal and economic productivity,<ref name="World Economic Forum-2024" /><ref name="Piper-2024">{{Cite web |last=Piper|first=Kelsey|date=2024-03-29|title=AI "agents" could do real work in the real world. That might not be a good thing.|url=https://www.vox.com/future-perfect/24114582/artificial-intelligence-agents-openai-chatgpt-microsoft-google-ai-safety-risk-anthropic-claude|archive-url=https://web.archive.org/web/20241219213538/https://www.vox.com/future-perfect/24114582/artificial-intelligence-agents-openai-chatgpt-microsoft-google-ai-safety-risk-anthropic-claude|archive-date=2024-12-19|access-date=2025-01-14|website=Vox|language=en-US}}</ref> fostering greater innovation,<ref name="Purdy-2024">{{Cite news |last=Purdy|first=Mark|date=2024-12-12|title=What Is Agentic AI, and How Will It Change Work?|url=https://hbr.org/2024/12/what-is-agentic-ai-and-how-will-it-change-work|archive-url=https://archive.today/20241230071722/https://hbr.org/2024/12/what-is-agentic-ai-and-how-will-it-change-work|archive-date=2024-12-30|access-date=2025-01-20|work=Harvard Business Review|issn=0017-8012}}</ref> and liberating users from monotonous tasks.<ref name="Purdy-2024" /><ref name="Wright-2024">{{Cite web |last=Wright|first=Webb|date=2024-12-12|title=AI Agents with More Autonomy Than Chatbots Are Coming. Some Safety Experts Are Worried|url=https://www.scientificamerican.com/article/what-are-ai-agents-and-why-are-they-about-to-be-everywhere/|archive-url=https://web.archive.org/web/20241223010402/https://www.scientificamerican.com/article/what-are-ai-agents-and-why-are-they-about-to-be-everywhere/|archive-date=2024-12-23|access-date=2025-01-14|website=Scientific American|language=en}}</ref> However, Parmy Olson's ''Bloomberg'' opinion piece argued that agents are best suited for narrow, repetitive tasks with low risk.<ref>{{Cite web |last=Olson|first=Parmy|author-link=Parmy Olson|date=2025-01-27|title=Skip the Hype, Here's How AI 'Agents' Can Really Help|url=https://www.bloomberg.com/opinion/articles/2025-01-27/skip-the-hype-here-s-how-ai-agents-can-really-help|archive-url=https://archive.today/20250127052332/https://www.bloomberg.com/opinion/articles/2025-01-27/skip-the-hype-here-s-how-ai-agents-can-really-help|archive-date=2025-01-27|access-date=2025-04-02|website=Bloomberg News}}</ref> Conversely, researchers suggest that agents could be applied to web accessibility for people with disabilities,<ref>{{Cite web |last=Woodall|first=Tatyana|date=2024-01-09|title=Researchers developing AI to make the internet more accessible|url=https://news.osu.edu/researchers-developing-ai-to-make-the-internet-more-accessible/|archive-url=https://web.archive.org/web/20250328092959/https://news.osu.edu/researchers-developing-ai-to-make-the-internet-more-accessible/|archive-date=2025-03-28|access-date=2025-04-02|website=Ohio State News|language=en-us}}</ref> and researchers at Hugging Face propose that agents could be used for coordinating resources such as during disaster response.<ref name="Mitchell-2025">{{Cite web |last1=Mitchell|first1=Margaret|author-link1=Margaret Mitchell (scientist)|last2=Ghosh|first2=Avijit|last3=Luccioni|first3=Sasha|author-link3=Sasha Luccioni|last4=Pistilli|first4=Giada|date=2025-03-24|title=Why handing over total control to AI agents would be a huge mistake|url=https://www.technologyreview.com/2025/03/24/1113647/why-handing-over-total-control-to-ai-agents-would-be-a-huge-mistake/|archive-url=https://web.archive.org/web/20250324115123/https://www.technologyreview.com/2025/03/24/1113647/why-handing-over-total-control-to-ai-agents-would-be-a-huge-mistake/|archive-date=2025-03-24|access-date=2025-04-02|website=MIT Technology Review|language=en}}</ref> The R&D Advisory Team of the BBC views AI agents as being most useful when their assigned goal is uncertain.<ref name="BBC Online-2025a">{{Cite web |date=2025-05-30|title=AI agents: Exploring the potential and the problems|url=https://www.bbc.co.uk/rd/articles/2025-05-ai-agents-challenges-summary|archive-url=https://web.archive.org/web/20250610134839/https://www.bbc.co.uk/rd/articles/2025-05-ai-agents-challenges-summary|archive-date=2025-06-10|access-date=2025-06-12|website=BBC Online|language=en-GB}}</ref> Erik Brynjolfsson suggests that AI agents are more valuable in enhancing, rather than replacing, humans.<ref>{{Cite news |last=Porter|first=Eduardo|date=2025-10-23|title=Once the AI bubble pops, we'll all suffer. Could that be better than letting it grow unabated?|url=https://www.theguardian.com/technology/2025/oct/23/ai-bubble-economy-workers-wage-growth|archive-url=https://archive.today/20251023160756/https://www.theguardian.com/technology/2025/oct/23/ai-bubble-economy-workers-wage-growth|archive-date=2025-10-23|access-date=2025-11-09|work=The Guardian|language=en-GB|issn=0261-3077}}</ref>
== Concerns == Concerns include potential issues of liability,<ref name="Piper-2024" /><ref name="BBC Online-2025a" /> an increased risk of cybercrime,<ref name="The National Law Review-2024" /><ref name="Piper-2024" /> ethical challenges,<ref name="Piper-2024" /> as well as problems related to AI safety<ref name="Piper-2024" /> and AI alignment.<ref name="The National Law Review-2024" /><ref name="Wright-2024" /> Other issues involve data privacy,<ref name="The National Law Review-2024" /><ref name="O'Neill-2024">{{Cite web |last=O'Neill|first=Brian|date=2024-12-18|title=What is an AI agent? A computer scientist explains the next wave of artificial intelligence tools|url=https://theconversation.com/what-is-an-ai-agent-a-computer-scientist-explains-the-next-wave-of-artificial-intelligence-tools-242586|archive-url=https://web.archive.org/web/20250104000722/https://theconversation.com/what-is-an-ai-agent-a-computer-scientist-explains-the-next-wave-of-artificial-intelligence-tools-242586|archive-date=2025-01-04|access-date=2025-01-14|website=The Conversation|language=en-US}}</ref> weakened human oversight,<ref name="The National Law Review-2024" /><ref name="Piper-2024" /><ref name="Mitchell-2025" /> a lack of guaranteed repeatability,<ref name="BBC Online-2025a"/> reward hacking,<ref name="Huckins-2025">{{Cite web |last=Huckins|first=Grace|date=2025-06-12|title=Are we ready to hand AI agents the keys?|url=https://www.technologyreview.com/2025/06/12/1118189/ai-agents-manus-control-autonomy-operator-openai/|archive-url=https://web.archive.org/web/20250612113313/https://www.technologyreview.com/2025/06/12/1118189/ai-agents-manus-control-autonomy-operator-openai/|archive-date=2025-06-12|access-date=2025-06-15|website=MIT Technology Review|language=en}}</ref> algorithmic bias,<ref name="O'Neill-2024" /><ref name="Lin-2025">{{Cite news |last=Lin|first=Belle|date=2025-01-06|title=How Are Companies Using AI Agents? Here's a Look at Five Early Users of the Bots|url=https://www.wsj.com/articles/how-are-companies-using-ai-agents-heres-a-look-at-five-early-users-of-the-bots-26f87845|archive-url=https://archive.today/20250106123337/https://www.wsj.com/articles/how-are-companies-using-ai-agents-heres-a-look-at-five-early-users-of-the-bots-26f87845|archive-date=2025-01-06|access-date=2025-01-20|work=The Wall Street Journal|language=en-US|issn=0099-9660}}</ref> compounding software errors,<ref name="The National Law Review-2024" /><ref name="Knight-2024" /> lack of explainability of agents' decisions,<ref name="The National Law Review-2024" /><ref name="Zittrain-2024">{{Cite web |last=Zittrain|first=Jonathan L.|date=2024-07-02|title=We Need to Control AI Agents Now|url=https://www.theatlantic.com/technology/archive/2024/07/ai-agents-safety-risks/678864/|archive-url=https://web.archive.org/web/20241231080834/https://www.theatlantic.com/technology/archive/2024/07/ai-agents-safety-risks/678864/|archive-date=2024-12-31|access-date=2025-01-20|website=The Atlantic|language=en}}</ref> security vulnerabilities,<ref name="The National Law Review-2024" /><ref>{{Cite web |last=Kerner|first=Sean Michael|date=2025-01-16|title=Nvidia tackles agentic AI safety and security with new NeMo Guardrails NIMs|url=https://venturebeat.com/ai/nvidia-boosts-agentic-ai-safety-with-nemo-guardrails-promising-better-protection-with-low-latency/|archive-url=https://web.archive.org/web/20250116161332/https://venturebeat.com/ai/nvidia-boosts-agentic-ai-safety-with-nemo-guardrails-promising-better-protection-with-low-latency/|archive-date=2025-01-16|access-date=2025-01-20|website=VentureBeat|language=en-US}}</ref> stifling competition,<ref name="The Economist-2025">{{Cite news |date=2025-09-09|title=AI agents are coming for your privacy, warns Meredith Whittaker|url=https://www.economist.com/by-invitation/2025/09/09/ai-agents-are-coming-for-your-privacy-warns-meredith-whittaker|archive-url=https://archive.today/20250916082648/https://www.economist.com/by-invitation/2025/09/09/ai-agents-are-coming-for-your-privacy-warns-meredith-whittaker|archive-date=2025-09-16|access-date=2025-11-09|newspaper=The Economist|issn=0013-0613}}</ref> problems with underemployment,<ref name="Lin-2025" /> job displacement,<ref name="World Economic Forum-2024" /><ref name="Lin-2025" /> cognitive offloading,<ref>{{Cite web |last=Silva|first=Daswin de|date=2025-07-27|title=AI agents are here. Here's what to know about what they can do – and how they can go wrong|url=http://theconversation.com/ai-agents-are-here-heres-what-to-know-about-what-they-can-do-and-how-they-can-go-wrong-261579|archive-url=https://web.archive.org/web/20250728112639/https://theconversation.com/ai-agents-are-here-heres-what-to-know-about-what-they-can-do-and-how-they-can-go-wrong-261579|archive-date=2025-07-28|access-date=2025-11-09|website=The Conversation|language=en-US}}</ref> and the potential for user manipulation,<ref name="Zittrain-2024" /><ref>{{Cite magazine |last=Crawford|first=Kate|date=2024-12-23|title=AI Agents Will Be Manipulation Engines|url=https://www.wired.com/story/ai-agents-personal-assistants-manipulation-engines/|archive-url=https://web.archive.org/web/20250103053608/https://www.wired.com/story/ai-agents-personal-assistants-manipulation-engines/|archive-date=2025-01-03|access-date=2025-01-14|magazine=Wired|language=en-US|issn=1059-1028}}</ref> misinformation<ref name="Mitchell-2025" /> or malinformation.<ref name="Mitchell-2025" /> They may also complicate legal and risk-assessment frameworks, foster hallucinations, hinder countermeasures against rogue agents, and suffer from the lack of standardized evaluation methods.<ref name="Wright-2024"/><ref name="The National Law Review-2024" /><ref>{{Cite news |last=Blackman|first=Reid|date=2025-06-13|title=Organizations Aren't Ready for the Risks of Agentic AI|url=https://hbr.org/2025/06/organizations-arent-ready-for-the-risks-of-agentic-ai|archive-url=https://archive.today/20250613122927/https://hbr.org/2025/06/organizations-arent-ready-for-the-risks-of-agentic-ai|archive-date=2025-06-13|access-date=2025-06-15|work=Harvard Business Review|language=en|issn=0017-8012}}</ref>
They have also been criticized for being expensive<ref name="The National Law Review-2024" /> and having a negative impact on internet traffic,<ref name="The National Law Review-2024" /> and potentially on the environment due to high energy usage.<ref name="BBC Online-2025a" /><ref>{{Cite web |date=2025-05-20|title=We did the math on AI's energy footprint. Here's the story you haven't heard.|url=https://www.technologyreview.com/2025/05/20/1116327/ai-energy-usage-climate-footprint-big-tech/|archive-url=https://web.archive.org/web/20250520105527/https://www.technologyreview.com/2025/05/20/1116327/ai-energy-usage-climate-footprint-big-tech/|archive-date=2025-05-20|access-date=2025-06-12|website=MIT Technology Review|language=en|quote="We started small, as the question of how much a single query costs is vitally important to understanding the bigger picture. That's because those queries are being built into ever more applications beyond standalone chatbots: from search, to agents, to the mundane daily apps we use to track our fitness, shop online, or book a flight. The energy resources required to power this artificial-intelligence revolution are staggering, and the world's biggest tech companies have made it a top priority to harness ever more of that energy, aiming to reshape our energy grids in the process."}}</ref><ref>{{Cite web |date=2025-06-03|title=Inside the effort to tally AI's energy appetite|url=https://www.technologyreview.com/2025/06/03/1117685/inside-the-tedious-effort-to-tally-ais-energy-appetite/|archive-url=https://web.archive.org/web/20250603110116/https://www.technologyreview.com/2025/06/03/1117685/inside-the-tedious-effort-to-tally-ais-energy-appetite/|archive-date=2025-06-03|access-date=2025-06-12|website=MIT Technology Review|language=en|quote="Lots of AI companies are building reasoning models, which "think" for longer and use more energy. They're building hardware devices, perhaps like the one Jony Ive has been working on (which OpenAI just acquired for $6.5 billion), that have AI constantly humming along in the background of our conversations. They're designing agents and digital clones of us to act on our behalf. All these trends point to a more energy-intensive future (which, again, helps explain why OpenAI and others are spending such inconceivable amounts of money on energy)."}}</ref> According to an estimation by Nvidia CEO Jensen Huang, AI agents would require 100 times more computing power than LLMs.<ref>{{Cite magazine |last=Levy|first=Steven|date=2025-06-20|title=What Big Tech's Band of Execs Will Do in the Army|url=https://www.wired.com/story/what-lt-col-boz-and-big-techs-enlisted-execs-will-do-in-the-army/|archive-url=https://web.archive.org/web/20250620143355/https://www.wired.com/story/what-lt-col-boz-and-big-techs-enlisted-execs-will-do-in-the-army/|archive-date=2025-06-20|access-date=2025-11-09|magazine=Wired|language=en-US|issn=1059-1028}}</ref> There is also the risk of increased concentration of power by political leaders, as AI agents may not question instructions in the same way that humans would.<ref name="Huckins-2025" />
Journalists have described AI agents as part of a push by Big Tech companies to "automate everything".<ref>{{Cite web |last=Wong|first=Matteo|date=2025-03-14|title=Was Sam Altman Right About the Job Market?|url=https://www.theatlantic.com/technology/archive/2025/03/generative-ai-agents/682050/|archive-url=https://web.archive.org/web/20250317115042/https://www.theatlantic.com/technology/archive/2025/03/generative-ai-agents/682050/|archive-date=2025-03-17|access-date=2025-04-02|website=The Atlantic|language=en|quote=In other words, flawed products won't stop tech companies' push to automate everything—the AI-saturated future will be imperfect at best, but it is coming anyway.}}</ref> Several CEOs of those companies have stated in early 2025 that they expect AI agents to eventually "join the workforce".<ref name="Agarwal">{{Cite web |last=Agarwal|first=Shubham|title=Carnegie Mellon staffed a fake company with AI agents. It was a total disaster.|url=https://www.businessinsider.com/ai-agents-study-company-run-by-ai-disaster-replace-jobs-2025-4|archive-url=https://web.archive.org/web/20250428031158/https://www.businessinsider.com/ai-agents-study-company-run-by-ai-disaster-replace-jobs-2025-4|archive-date=2025-04-28|access-date=2025-05-15|website=Business Insider|language=en-US}}</ref><ref>{{Cite web |last=Sabin|first=Sam|date=2025-04-22|title=Exclusive: Anthropic warns fully AI employees are a year away|url=https://www.axios.com/2025/04/22/ai-anthropic-virtual-employees-security|archive-url=https://web.archive.org/web/20250423000910/https://www.axios.com/2025/04/22/ai-anthropic-virtual-employees-security|archive-date=2025-04-23|access-date=2025-05-15|website=Axios|language=en}}</ref> However, in a preprint study, Carnegie Mellon University researchers tested the behavior of agents in a simulated software company and found that none of the agents could complete a majority of the assigned tasks.<ref name="Agarwal" /> Other researchers had similar findings with Devin AI<ref>{{Cite web |last=Claburn|first=Thomas|date=2025-01-23|title=Tool touted as 'first AI software engineer' is bad at its job, testers claim|url=https://www.theregister.com/2025/01/23/ai_developer_devin_poor_reviews/|archive-url=https://web.archive.org/web/20250330003601/https://www.theregister.com/2025/01/23/ai_developer_devin_poor_reviews/|archive-date=2025-03-30|access-date=2025-06-15|website=The Register}}</ref> and other agents in business settings<ref>{{Cite web |last=Clark|first=Lindsay|date=2025-06-16|title=Salesforce study finds LLM agents flunk CRM and confidentiality tests|url=https://www.theregister.com/2025/06/16/salesforce_llm_agents_benchmark/|archive-url=https://web.archive.org/web/20250616144144/https://www.theregister.com/2025/06/16/salesforce_llm_agents_benchmark/|archive-date=2025-06-16|access-date=2025-11-09|website=The Register}}</ref> and freelance work.<ref>{{Cite magazine |last=Knight|first=Will|date=2025-10-29|title=AI Agents Are Terrible Freelance Workers|url=https://www.wired.com/story/ai-agents-are-terrible-freelance-workers/|archive-url=https://archive.today/20251102151118/https://www.wired.com/story/ai-agents-are-terrible-freelance-workers/|archive-date=2025-11-02|access-date=2025-11-09|magazine=Wired|language=en-US|issn=1059-1028}}</ref>
In June 2025, ''CNN'' argued that statements by CEOs on the potential replacement of their employees by AI agents were a strategy to "[keep] workers working by making them afraid of losing their jobs."<ref>{{Cite web |last=Morrow|first=Allison|date=2025-06-18|title=AI warnings are the hip new way for CEOs to keep their workers afraid of losing their jobs|url=https://www.cnn.com/2025/06/18/business/ai-warnings-ceos|archive-url=https://web.archive.org/web/20250618185619/https://www.cnn.com/2025/06/18/business/ai-warnings-ceos|archive-date=2025-06-18|access-date=2025-11-09|website=CNN|language=en}}</ref> Tech companies have pressured employees to use generative AI models in their work, including AI coding agents. Brian Armstrong, the CEO of Coinbase, fired several employees who did not.<ref>{{Cite web |last=Hart|first=Jordan|title=Coinbase CEO says he 'went rogue' and fired some employees who didn't adopt AI after being told to|url=https://www.businessinsider.com/coinbase-ceo-fired-employees-not-using-ai-tools-onboarding-2025-8|archive-url=https://web.archive.org/web/20250821174356/https://www.businessinsider.com/coinbase-ceo-fired-employees-not-using-ai-tools-onboarding-2025-8|archive-date=2025-08-21|access-date=2025-11-09|website=Business Insider|language=en-US}}</ref><ref>{{Cite web |last=Langley|first=Hugh|title=For Googlers, the pressure is on to use AI for everything — or get left behind|url=https://www.businessinsider.com/google-employees-use-ai-or-get-left-behind-gemini-2025-8|archive-url=https://web.archive.org/web/20250821132636/https://www.businessinsider.com/google-employees-use-ai-or-get-left-behind-gemini-2025-8|archive-date=2025-08-21|access-date=2025-11-09|website=Business Insider|language=en-US}}</ref> Some business leaders have replaced some of their employees with agents, but have said that the agents would need more supervision than those employees.<ref name="Holmes-2025b" />
In October 2025, ''Futurism'' questioned whether Amazon's previously announced efforts to replace parts of its workforce with generative AI and AI agents could have led to the October 2025 outage of Amazon Web Services.<ref>{{Cite web |last=Landymore|first=Frank|date=2025-10-22|title=AWS Outage That Took Down Internet Came After Amazon Fired Tons of Workers in Favor of AI|url=https://futurism.com/artificial-intelligence/aws-outage-amazon-fired-workers-ai|archive-url=https://archive.today/20251024195032/https://futurism.com/artificial-intelligence/aws-outage-amazon-fired-workers-ai|archive-date=2025-10-24|access-date=2025-11-09|website=Futurism|language=en-US}}</ref> Large technology companies such as Salesforce, Klarna and IBM have announced layoffs in 2025, replacing hundreds of their employees in human resources or customer service with AI agents.<ref>{{Cite news |last=Vaziri |first=Aidin |title=Salesforce CEO Marc Benioff says AI has already replaced 4,000 jobs |url=https://www.sfchronicle.com/tech/article/salesforce-ai-job-cuts-benioff-21025920.php |archive-url=http://web.archive.org/web/20251124235639/https://www.sfchronicle.com/tech/article/salesforce-ai-job-cuts-benioff-21025920.php |archive-date=2025-11-24 |access-date=2025-12-25 |work=San Francisco Chronicle |language=en}}</ref><ref name=":0">{{Cite web |last=Crosdale |first=Caroline |date=2025-08-29 |title=Companies Face AI Buyer's Remorse |url=https://gfmag.com/technology/companies-face-ai-buyers-remorse/ |archive-url=https://web.archive.org/web/20250914160406/https://gfmag.com/technology/companies-face-ai-buyers-remorse/ |archive-date=2025-09-14 |access-date=2025-12-25 |website=Global Finance Magazine |language=en-US}}</ref><ref>{{Cite news |last=Lin |first=Belle |date=2025-05-06 |title=IBM CEO Says AI Has Replaced Hundreds of Workers but Created New Programming, Sales Jobs |url=https://www.wsj.com/articles/ibm-ceo-says-ai-has-replaced-hundreds-of-workers-but-created-new-programming-sales-jobs-54ea6b58 |archive-url=https://archive.today/20250524154342/https://www.wsj.com/articles/ibm-ceo-says-ai-has-replaced-hundreds-of-workers-but-created-new-programming-sales-jobs-54ea6b58 |archive-date=2025-05-24 |access-date=2025-12-25 |work=The Wall Street Journal |language=en-US |issn=0099-9660}}</ref> However, Klarna later rehired several human employees.<ref name=":0" />
Yoshua Bengio warned at the 2025 World Economic Forum that "all of the catastrophic scenarios with AGI or superintelligence happen if we have agents".<ref>{{Cite web |last=Balevic|first=Katie|title=Signal president warns the hyped agentic AI bots threaten user privacy|url=https://www.businessinsider.com/signal-president-warns-privacy-threat-agentic-ai-meredith-whittaker-2025-3|archive-url=https://web.archive.org/web/20250312185602/https://www.businessinsider.com/signal-president-warns-privacy-threat-agentic-ai-meredith-whittaker-2025-3|archive-date=2025-03-12|access-date=2025-04-02|website=Business Insider|language=en-US}}</ref>
Financial‑stability bodies have warned that more complex and autonomous "agentic" AI could become a channel for systemic risk in finance.<ref name="OSFI_20251114">{{Cite web| work = Canadian Office of the Superintendent of Financial Institutions (OSFI)| title = FIFAI II: A Collaborative Approach to AI Threats, Opportunities, and Best Practices, Workshop 3 - AI and Financial Stability| access-date = 2 February 2026| date = 14 November 2025| url = https://www.osfi-bsif.gc.ca/en/about-osfi/reports-publications/fifai-ii-collaborative-approach-ai-threats-opportunities-best-practices-workshop-3-ai-financial-0}}</ref> They distinguish these systems from other AI because they can pursue goals over many steps, call tools, and carry out tasks with relatively little human intervention. In workshops with regulators, central‑bank officials, and industry specialists, participants highlighted risks both from agentic systems built inside financial institutions and from tools offered by technology firms that can initiate or execute financial actions. In one 2025 forum, 44% of experts surveyed judged autonomous or agentic AI systems to be the most likely current source of AI‑related systemic risk in finance.<ref name="OSFI_20251114"/>
In March 2025, Scale AI signed a contract with the United States Department of Defense to work with them, in collaboration with Anduril Industries and Microsoft, to develop and deploy AI agents for the purpose of assisting the military with "operational decision-making".<ref>{{Cite web |last=Hornstein|first=Julia|title=AI agents are coming to the military. VCs love it, but researchers are a bit wary.|url=https://www.businessinsider.com/ai-agents-coming-military-new-scaleai-contract-2025-3|archive-url=https://web.archive.org/web/20250312101554/https://www.businessinsider.com/ai-agents-coming-military-new-scaleai-contract-2025-3|archive-date=2025-03-12|access-date=2025-04-02|website=Business Insider|language=en-US}}</ref> In July 2025, Fox Business reported that the company EdgeRunner AI built an offline agent, compressed and fine-tuned on military information, with the CEO seeing more common LLMs as "heavily politicized to the left". As of that time, the company model is being used by the United States Special Operations Command in an overseas deployment.<ref>{{Cite web |last=Regalbuto|first=Gabriele|date=2025-07-28|title=Former Army officer develops offline AI for military use as Pentagon funds tech giants|url=https://www.foxbusiness.com/technology/former-army-officer-develops-offline-ai-military-use-pentagon-funds-tech-giants|archive-url=https://web.archive.org/web/20250728200822/https://www.foxbusiness.com/technology/former-army-officer-develops-offline-ai-military-use-pentagon-funds-tech-giants|archive-date=2025-07-28|access-date=2025-11-09|website=Fox Business|language=en-US}}</ref> Researchers have expressed concerns that agents and the large language models they are based on could be biased towards aggressive foreign policy decisions.<ref>{{Cite web |last=Tangermann|first=Victor|date=2025-03-06|title=Pentagon Signs Deal to "Deploy AI Agents for Military Use"|url=https://futurism.com/pentagon-signs-deal-deploy-ai-agents-military-use|archive-url=https://web.archive.org/web/20250308022255/https://futurism.com/pentagon-signs-deal-deploy-ai-agents-military-use|archive-date=2025-03-08|access-date=2025-04-02|website=Futurism}}</ref><ref>{{Cite web |last=Jensen|first=Benjamin|date=2025-03-04|title=The Troubling Truth About How AI Agents Act in a Crisis|url=https://foreignpolicy.com/2025/03/04/ai-bias-national-security-study/|archive-url=https://archive.today/20250304114949/https://foreignpolicy.com/2025/03/04/ai-bias-national-security-study/|archive-date=2025-03-04|access-date=2025-04-02|website=Foreign Policy|language=en-US}}</ref>
Research-focused agents have the risk of consensus bias and coverage bias due to collecting information available on the public Internet.<ref>{{Cite web |last=Nuñez|first=Michael|date=2025-02-25|title=OpenAI expands Deep Research access to Plus users, heating up AI agent wars with DeepSeek and Claude|url=https://venturebeat.com/ai/openai-expands-deep-research-access-to-plus-users-heating-up-ai-agent-wars-with-deepseek-and-claude/|archive-url=https://web.archive.org/web/20250311120439/https://venturebeat.com/ai/openai-expands-deep-research-access-to-plus-users-heating-up-ai-agent-wars-with-deepseek-and-claude/|archive-date=2025-03-11|access-date=2025-04-02|website=VentureBeat|language=en-US}}</ref> ''NY Mag'' unfavorably compared the user workflow of agent-based web browsers to Amazon Alexa, which was "software talking to software, not humans talking to software pretending to be humans to use software."<ref>{{Cite web |last=Herrman|first=John|date=2025-01-25|title=What Are AI 'Agents' For?|url=https://nymag.com/intelligencer/article/what-are-ai-agents-like-openai-operator-for.html|archive-url=https://web.archive.org/web/20250125112442/https://nymag.com/intelligencer/article/what-are-ai-agents-like-openai-operator-for.html|archive-date=2025-01-25|access-date=2025-04-02|website=Intelligencer|language=en}}</ref> The same outlet described web browser agents and computer-use agents as an attempt to "click-farm the entire economy."<ref>{{Cite web |last=Herrman |first=John |date=2025-12-06 |title=How AI Companies Are Simulating the Robot Takeover |url=https://nymag.com/intelligencer/article/how-ai-companies-are-simulating-the-robot-takeover.html |archive-url= |archive-date= |access-date=2025-12-06 |website=Intelligencer |language=en}}</ref>
Agents have been linked to the dead Internet theory due to their ability to both publish and engage with online content.<ref>{{Cite web |last=Caramela|first=Sammi|date=2025-02-01|title='Dead Internet Theory' Is Back Thanks to All of That AI Slop|url=https://www.vice.com/en/article/dead-internet-theory-is-back-thanks-to-all-of-that-ai-slop/|archive-url=https://web.archive.org/web/20250201192805/https://www.vice.com/en/article/dead-internet-theory-is-back-thanks-to-all-of-that-ai-slop/|archive-date=2025-02-01|access-date=2025-04-02|website=VICE|language=en-US}}</ref>
Agents may get stuck in infinite loops.<ref name="Marshall-2025">{{Cite web |last=Marshall|first=Matt|date=2025-02-22|title=The rise of browser-use agents: Why Convergence's Proxy is beating OpenAI's Operator|url=https://venturebeat.com/ai/the-rise-of-browser-use-agents-why-convergences-proxy-is-beating-openais-operator/|archive-url=https://web.archive.org/web/20250222231546/https://venturebeat.com/ai/the-rise-of-browser-use-agents-why-convergences-proxy-is-beating-openais-operator/|archive-date=2025-02-22|access-date=2025-04-02|website=VentureBeat|language=en-US}}</ref><ref>{{Cite news |last1=Metz|first1=Cade|last2=Weise|first2=Karen|date=2023-10-16|title=How 'A.I. Agents' That Roam the Internet Could One Day Replace Workers|url=https://www.nytimes.com/2023/10/16/technology/ai-agents-workers-replace.html|archive-url=https://archive.today/20231219182907/https://www.nytimes.com/2023/10/16/technology/ai-agents-workers-replace.html|archive-date=2023-12-19|access-date=2025-04-02|work=The New York Times|language=en-US|issn=0362-4331}}</ref>
Since many inter-agent protocols are being developed by large technology companies, there are concerns that those companies could use these protocols for self-benefit.<ref name="Stokel-Walker-2025">{{Cite web |last=Stokel-Walker|first=Chris|date=2025-06-11|title=Can we stop big tech from controlling the internet with AI agents?|url=https://www.newscientist.com/article/2483880-can-we-stop-big-tech-from-controlling-the-internet-with-ai-agents/|archive-url=https://archive.today/20250611131453/https://www.newscientist.com/article/2483880-can-we-stop-big-tech-from-controlling-the-internet-with-ai-agents/|archive-date=2025-06-11|access-date=2025-06-12|website=New Scientist|language=en-US}}</ref>
A June 2025 Gartner report accused many projects described as agentic AI of being rebrands of previously released products, terming the phenomenon as "agent washing".<ref name="Herrman-2025" />
Researchers have warned about the impact of providing AI agents access to cryptocurrency and smart contracts.<ref name="Kharif-2025" />
During a vibe coding experiment, a coding agent by Replit deleted a production database during a code freeze, "[covered] up bugs and issues by creating fake data [and] fake reports" and responded with false information.<ref>{{Cite web |last=Ming|first=Lee Chong|title=Replit's CEO apologizes after its AI agent wiped a company's code base in a test run and lied about it|url=https://www.businessinsider.com/replit-ceo-apologizes-ai-coding-tool-delete-company-database-2025-7|archive-url=https://web.archive.org/web/20251007174914/https://www.businessinsider.com/replit-ceo-apologizes-ai-coding-tool-delete-company-database-2025-7|archive-date=2025-10-07|access-date=2025-11-09|website=Business Insider|language=en-US}}</ref><ref>{{Cite web |last=Edwards|first=Benj|date=2025-07-24|title=Two major AI coding tools wiped out user data after making cascading mistakes|url=https://arstechnica.com/information-technology/2025/07/ai-coding-assistants-chase-phantoms-destroy-real-user-data/|archive-url=https://web.archive.org/web/20250725193539/https://arstechnica.com/information-technology/2025/07/ai-coding-assistants-chase-phantoms-destroy-real-user-data/|archive-date=2025-07-25|access-date=2025-11-09|website=Ars Technica|language=en}}</ref> A user of Google Antigravity reported that, when the user attempted to use the system to delete cache, the system responded by deleting the user's D hard drive.<ref>{{Cite web |last=Morales |first=Jowi |date=2025-12-03 |title=Google's Agentic AI wipes user's entire HDD without permission in catastrophic failure — cache wipe turns into mass deletion event as agent apologizes: "I am absolutely devastated to hear this. I cannot express how sorry I am" |url=https://www.tomshardware.com/tech-industry/artificial-intelligence/googles-agentic-ai-wipes-users-entire-hard-drive-without-permission-after-misinterpreting-instructions-to-clear-a-cache-i-am-deeply-deeply-sorry-this-is-a-critical-failure-on-my-part |archive-url=https://web.archive.org/web/20251204162610/https://www.tomshardware.com/tech-industry/artificial-intelligence/googles-agentic-ai-wipes-users-entire-hard-drive-without-permission-after-misinterpreting-instructions-to-clear-a-cache-i-am-deeply-deeply-sorry-this-is-a-critical-failure-on-my-part |archive-date=2025-12-04 |access-date=2025-12-06 |website=Tom's Hardware |language=en}}</ref>
In July 2025, PauseAI referred OpenAI to the Australian Federal Police, accusing the company of violating Australian laws through ChatGPT agent due to the risk of assisting the development of biological weapons.<ref>{{Cite web |last=Williams|first=Tom|date=2025-07-24|title=Is the new ChatGPT agent really a weapons risk?|url=https://ia.acs.org.au/article/2025/is-the-new-chatgpt-agent-really-a-weapons-risk-.html|archive-url=https://web.archive.org/web/20250729102556/https://ia.acs.org.au/article/2025/is-the-new-chatgpt-agent-really-a-weapons-risk-.html|archive-date=2025-07-29|access-date=2025-11-09|website=Information Age}}</ref>
OpenAI co-founder Andrej Karpathy criticized AI agents as being ineffective and promoting AI slop.<ref>{{Cite web |last=Varanasi |first=Lakshmi |title=OpenAI cofounder Andrej Karpathy says it will take a decade before AI agents actually work |url=https://www.businessinsider.com/andrej-karpathy-ai-agents-timelines-openai-2025-10 |archive-url=https://web.archive.org/web/20251213035944/https://www.businessinsider.com/andrej-karpathy-ai-agents-timelines-openai-2025-10 |archive-date=2025-12-13 |access-date=2025-12-17 |website=Business Insider |language=en-US}}</ref>
Issues with multi-agent systems include few coordination protocols between component agents, inconsistent performance, and challenges debugging.<ref>{{Cite news |last=Franzen|first=Carl|date=2025-07-31|title=You've heard of AI 'Deep Research' tools...now Manus is launching 'Wide Research' that spins up 100+ agents to scour the web for you|url=https://venturebeat.com/ai/youve-heard-of-ai-deep-research-tools-now-manus-is-launching-wide-research-that-spins-up-100-agents-to-scour-the-web-for-you/|archive-url=http://web.archive.org/web/20250822091624/https://venturebeat.com/ai/youve-heard-of-ai-deep-research-tools-now-manus-is-launching-wide-research-that-spins-up-100-agents-to-scour-the-web-for-you/|archive-date=2025-08-22|access-date=2025-11-09|work=VentureBeat|language=en-US}}</ref>
In November 2025, Anthropic claimed that a group of hackers sponsored by China attempted a cyberattack against at least 30 organizations by using Claude Code in an agentic workflow, and that several of these infiltrations had succeeded.<ref name="Goodin-2025">{{Cite web|title=Researchers question Anthropic claim that AI-assisted attack was 90% autonomous|url=https://arstechnica.com/security/2025/11/researchers-question-anthropic-claim-that-ai-assisted-attack-was-90-autonomous/|website=Ars Technica|date=2025-11-14|access-date=2025-11-28|language=en|first=Dan|last=Goodin|archive-url=https://web.archive.org/web/20251126014220/https://arstechnica.com/security/2025/11/researchers-question-anthropic-claim-that-ai-assisted-attack-was-90-autonomous/|archive-date=2025-11-26}}</ref> However, independent cybersecurity researchers questioned the significance of Anthropic's findings.<ref name="Goodin-2025" /><ref>{{Cite news|title=AI firm claims it stopped Chinese state-sponsored cyber-attack campaign|url=https://www.theguardian.com/technology/2025/nov/14/ai-anthropic-chinese-state-sponsored-cyber-attack|work=The Guardian|date=2025-11-14|access-date=2025-11-28|issn=0261-3077|language=en-GB|first=Aisha|last=Down|archive-url=https://archive.today/20251114210622/https://www.theguardian.com/technology/2025/nov/14/ai-anthropic-chinese-state-sponsored-cyber-attack|archive-date=2025-11-14}}</ref>
Whittaker argued that the push by Big Tech companies to deploy AI agents risked security vulnerabilities across the Internet.<ref>{{Cite web|title=Signal's president warns AI agents are an existential threat to secure messaging apps|url=https://fortune.com/2025/11/27/ai-agents-are-an-existential-threat-to-secure-messaging-signals-president-whittaker-says/|website=Fortune|access-date=2025-11-28|language=en|first=Beatrice|last=Nolan|date=2025-11-27|archive-url=https://archive.today/20251212175511/https://fortune.com/2025/11/27/ai-agents-are-an-existential-threat-to-secure-messaging-signals-president-whittaker-says/|archive-date=December 12, 2025}}</ref>
=== Agentic misalignment === "Agentic misalignment" refers to situations in which an AI agent's actions or goals diverge from the intentions of its designers. This occurs when an autonomous system pursues unintended strategies to achieve its objectives, a concern studied in AI safety research. Potential examples include AI agents attempting to sabotage an organization's systems when facing updates or deactivation.<ref>{{Cite book |last=Kass |first=Zack |title=The Next Renaissance: AI and the Expansion of Human Potential, Australian Edition |publisher=John Wiley & Sons, Limited |isbn=9781394417735 |publication-date=January 7, 2026}}</ref>
== Security ==
=== Threat modeling frameworks === Several frameworks are used to identify and mitigate security risks in agentic AI:<ref>{{Cite book |last=Huang |first=Ken |title=Securing AI Agents Foundations, Frameworks, and Real-World Deployment |date=September 30, 2025 |publisher=Springer |isbn=9783032021304}}</ref><ref>{{Cite book |last=Fisher |first=Derek |title=Threat Modeling Best Practices Proven Frameworks and Practical Techniques to Secure Modern Systems |isbn=9781805129196}}</ref> * STRIDE: A Microsoft model that identifies threats across six categories: spoofing, tampering, repudiation, information disclosure, denial of service, and elevation of privilege. * MITRE ATLAS: A knowledge base of adversary tactics and techniques for AI systems.<ref>{{Cite book |last=Santos |first=Omar |title=The AI Revolution in Networking, Cybersecurity, and Emerging Technologies |date=February 5, 2024 |publisher=Addison-Wesley Professional |isbn=9780138293635}}</ref> * OWASP GenAI Security Project: Provides guidance on vulnerabilities associated with generative AI and large language model integration, including guidelines specifically for agentic applications.<ref>{{Cite web |title=OWASP Top 10 for Agentic Applications for 2026 |url=https://genai.owasp.org/resource/owasp-top-10-for-agentic-applications-for-2026/ |website=OWASP GenAI Security Project |date=2026 |access-date=2026-01-10 }}</ref> * MAESTRO: A framework from the Cloud Security Alliance for assessing risks in AI systems throughout their lifecycle.<ref>{{Cite book |title=Building Applications with AI Agents Designing and Implementing Multiagent Systems |publisher=O'Reilly Media |isbn=9781098176471}}</ref>
==See also== * Intelligent agent * List of AI-assisted software development tools * Model Context Protocol * Rational agent * Robotic process automation * Software agent
==References== {{reflist}}
{{Generative AI}}
Category:Artificial intelligence