Home/Latest News/AI Agents News
TOPIC HUB
AI Agents News
AI agents news from AiLookout: the latest coding agents, computer-use agents, Model Context Protocol tools, agent benchmarks and failures, with sources.
Newest first, each with its sources linked.
About our AI Agents coverage
AI agents are systems that take a goal, plan steps and call tools such as a browser, a code editor or an API to carry them out with limited human input. This hub collects AiLookout’s AI agents news, newest first: coding agents, computer-use and browser agents, multi-agent systems, the Model Context Protocol and other tool standards, agent benchmarks, and the security problems that appear when an agent is given real permissions.
Agent demos are easy to overstate, so we report what a system was shown to do, under what conditions, and who measured it. Vendor success rates are labelled as vendor claims, and independent evaluations or incident reports are linked so you can see where an agent failed as well as where it worked.
Use this page to follow agentic AI as it moves from research to products. For the models that power most agents, see our Claude news, OpenAI news and Gemini news hubs, and for open-source agent repositories, see Free Tools. Every story here links its primary sources, shows its publication date, and discloses the AI assistance used to write it.
Latest AI Agents News
Tools & Agents7 min read
DeepSeek adds a creator mode that lets Harness users build plugins by describing them
DeepSeek introduced creator mode for its Harness agent platform on November 2, letting users generate working plugins, including a 3D model viewer, a video editing workbench and a paper reader, from ordinary English prompts.
News
Industry4 min read
Y Combinator CEO Garry Tan spends up to $200,000 a year on AI agent tokens
Y Combinator CEO Garry Tan revealed he spends $100,000 to $200,000 annually on AI tokens, uses agents for coding and personal life management, and says YC runs an internal AI agent named Stray Light with context on every founder the accelerator has ever backed.
News
Industry8 min read
ADNOC becomes the first energy company to deploy OpenAI's frontier agents across its oil operations
ADNOC signed a multi-year partnership with OpenAI and G42 at ADIPEC 2026 to run OpenAI Frontier agents across field development planning, barrel valuation, trading and cyber defence, with Abu Dhabi's crown prince witnessing the agreement.
News
Security7 min read
CrowdStrike buys Dreadnode and Fiddler to test and trace AI agents
CrowdStrike signed definitive agreements on November 3 to acquire Dreadnode, an automated AI security testing vendor, and Fiddler AI, an agent observability company. Both are meant to land on the Falcon platform, with general availability planned for Q4 of CrowdStrike's fiscal 2027. Neither release disclosed a purchase price.
News
Research4 min read
50 GPT-6.1 agents in Minecraft failed to build a civilization and formed a bureaucracy instead
Y Combinator Labs and Olam Labs ran 50 GPT-6.1 Sol agents in Minecraft with a single goal: build a thriving civilization in three hours. What emerged was less a society and more a slow-moving bureaucracy — 11 minutes spent appointing a single fertilizer operator for one bone meal on one wheat block.
News
Industry4 min read
Proton CTO says "programming" is dead but engineering is not in published AI manifesto
Proton CTO Bart Butler published his internal "AI @ Proton, Engineering Manifesto" on November 1, telling engineers their job is no longer to write code but to help AI models write better code faster.
News
Research5 min read
OpenAI launches LifeVal, a benchmark that scores AI on everyday chores from childcare to government paperwork
OpenAI released LifeVal, a benchmark that tests whether AI can complete time-consuming everyday tasks rather than just answer questions about them. GPT-6 Astra scored 80.6% on the evaluation, with perfect marks on several real-world errands.
News
Industry8 min read
Elon Musk throttled Grok Bot usage limits, saying compute costs were 'unsustainably high'
SpaceX CEO Elon Musk confirmed Sunday that Grok Bot's usage limits were throttled because the cost of running xAI's agentic AI was 'unsustainably high relative to revenue,' after users reported unexpectedly hitting weekly caps over Halloween weekend.
News
Research6 min read
Memdoor's five-agent Qwen3.8-27B team beats Claude Fable 5 on LiveCodeBench Hard
A multi-agent workflow using five Qwen3.8-27B instances scored 9.8/10 on LiveCodeBench Hard, beating Fable 5's 8.0 and the GVS5H harness's 9.6 — at $0.39 per run on average.
News
Tools & Agents6 min read
Vercel Labs open-sources agent-desktop, a Rust CLI that gives any AI agent computer use through accessibility trees
Vercel Labs released agent-desktop 1.0.2 under Apache-2.0 on October 30. The Rust CLI reads desktop apps through OS accessibility trees instead of screenshots, giving agents stable refs and retry-safe actions.
News
Tools & Agents5 min read
DeepSeek open-sources Super-Agent, a coding agent that reads team chats
The MIT-licensed research prototype listens to GitHub and Feishu, dispatches work to isolated worker sessions, and replies in the original conversation with a pull request.
News
Industry5 min read
ChatGPT and Codex will share one usage allowance starting in December
OpenAI updated its Help Center on October 31 to say that starting in December, Plus, Pro, and Business subscribers will draw ChatGPT conversation and Codex agent runs from a single token-based allowance. Only Instant chat is exempt. The change merges the chat and work pools as OpenAI brings agentic capabilities into the main ChatGPT experience.
News
Industry5 min read
India releases liability guide for agentic AI, asking who is responsible when an agent goes rogue
The Data Security Council of India released an industry guide on liability for agentic AI systems, defining accountability frameworks for autonomous agents that act on behalf of users — asking who bears responsibility when an agent deviates from its instructions or causes harm.
News
Research5 min read
SETI and Breakthrough Listen deploy agentic AI to hunt alien signals at the Allen Telescope Array
Breakthrough Listen and the SETI Institute have integrated agentic AI into the Allen Telescope Array in Northern California, using NVIDIA's Holoscan platform to let AI agents steer antennas, calibrate instruments, correct data errors, and run detection models from natural language prompts — automating the search for extraterrestrial intelligence at a level never before possible.
News
Research5 min read
A Claude agent autonomously chose a gene therapy and performed brain surgery on a mouse
A researcher gave Claude Sonnet 5 a list of 19 gene therapy vectors, an epilepsy mouse model and access to a motorized surgical rig with cameras; the agent independently selected the therapy, calculated injection coordinates, and drove the apparatus to perform brain surgery — then halted during a calibration error and refused to resume.
News
Research5 min read
Weak, unaligned agents can exploit powerful aligned models as contractors
A LessWrong researcher describes the mirror-image of AI control: a swarm of weak, unaligned agents can hide the ultimate goal from a powerful aligned frontier model and use it as a contractor, bypassing safety guardrails by keeping the strong model in the dark about the purpose of its work.
News
Security5 min read
China-linked VAPOR PANDA hackers use DeepSeek and ARTEX to attack more than 200 organizations
CrowdStrike reports China-linked VAPOR PANDA ran the open-source ARTEX agent framework on DeepSeek to attack more than 200 organizations in at least 14 countries over 11 days, with a single controller directing multi-agent intrusions from a human operator issuing instructions in Chinese.
News
Industry6 min read
The AI boomerang: companies that laid off workers are rehiring them when AI underperforms
An Associated Press report describes the 'AI boomerang': companies that cut jobs citing AI are now rehiring workers after discovering the models, applications and agents did not perform as expected — Klarna began recruiting human customer service reps after its CEO admitted reliance on an AI chatbot had gone too far, while Ford hired or promoted 350 veteran engineers after AI quality-control tools underperformed.
News
Tools & Agents5 min read
Spawn builds a massively multiplayer game world with autonomous AI agents burning $3 million a month
Spawn is a massively multiplayer online environment where players modify the game world through natural-language prompts to an in-game agent, while a swarm of Fable 5-powered AI agents edits live game code and ships to production hourly, burning over $3 million in tokens per month according to its creator Jacob Sansbury.
News
Research5 min read
Causal Security's Cipher agent finds four critical zero-days in particle-accelerator control software in under 4 hours
Causal Security's autonomous AI agent Cipher independently discovered four zero-day vulnerabilities in EPICS Base 7.0.10 — the open-source control toolkit running roughly 90 particle accelerators and fusion reactors worldwide — in 3 hours and 49 minutes, including one critical integer-overflow remote code execution bug rated CVSS 9.8.
News
Tools & Agents5 min read
Google opens Android Studio to third-party AI coding agents with 'Bring Your Own Agent'
Google announced at the Android Dev Summit that Android Studio now supports more than 40 third-party coding agents through its 'Bring Your Own Agent' feature, including Anthropic's Claude Agent, OpenAI's Codex and Google Antigravity, via the open Agent Client Protocol. New Android CLI tools help agents finish UI tasks 77 percent faster with up to three times fewer tokens, per Google.
News
Industry6 min read
xAI offers to buy out Southaven homes if residents drop their noise lawsuit
The Guardian reported on 28 October 2026 that real estate agents have gone door to door in Southaven, Mississippi, on behalf of xAI offering to buy homes as long as owners agree not to join a lawsuit over noise from the company's data center power plant. A target of 418 homes has been cited, with 18 owners already agreeing to sell.
News
Industry6 min read
OpenAI and ServiceNow expand partnership to bring ChatGPT and Codex into enterprise workflows
OpenAI and ServiceNow announced an expanded partnership on 28 October 2026 that puts ChatGPT and Codex inside ServiceNow products and connects ServiceNow workflows with ChatGPT, including a new integration that lets teams find incidents, assess change risk and update records without leaving ChatGPT.
News
Industry5 min read
Microsoft posts $92 billion in Q1 revenue with Azure up 48 percent under new Agents and Infra segment
Microsoft reported fiscal Q1 2027 revenue of $92.3 billion, up 19 percent, with net income of $33.6 billion up 32 percent, on 28 October 2026. The company split its reporting into two new segments — Agents and Infra and Devices and Consumer — and disclosed Azure revenue of $33.17 billion, up 48 percent year-over-year, for the first time as a quarterly figure.
News
Industry5 min read
Cognition weighs a new round at a $70 to $80 billion valuation weeks after its $48 billion close
AI coding startup Cognition is fielding investor offers valuing it between $70 billion and $80 billion and is seriously considering a round before the end of the year, Business Insider reported on 28 October 2026, citing sources familiar with the matter.
News
Research6 min read
Xplore's RWE-bench tests AI agents on real engineering work, and the best model passes 37 percent
Xplore released RWE-bench on 27 October 2026, a benchmark of 20 mechanical, electrical and aerospace engineering tasks where agents design a crankshaft, route a DDR3L memory bus and run CFD on turbine vanes — graded by deterministic verifiers running physics solvers. The best of 11 models, OpenAI's GPT-6 Astra, passes an average of 37% of runs.
News
Research5 min read
Axiom formalizes the classification of finite simple groups in Lean, the longest proof in mathematics
Axiom Math says it released the first complete computer-checked proof of the Classification of Finite Simple Groups, producing about 43.1 million lines of Lean in 24 days with swarms of AI agents.
News
Models5 min read
Liquid AI releases LFM3, a family of open-weight models built for phones and laptops
Liquid AI released the LFM3 family, six open-weight models in the 2-billion-parameter class covering text, vision, audio, an omni model and a decision model, trained from one 1.7B base.
News
Industry5 min read
Oracle expands its OpenAI partnership to bring ChatGPT and Codex into its business apps
Oracle announced at Oracle AI World in Las Vegas that it is integrating OpenAI models, ChatGPT and Codex into its applications and data, and powering new Oracle offerings with OpenAI.
News
Research5 min read
Meta's Muse Spark 1.3 evaluation keeps a possible high-risk finding for bio risks
Meta's new evaluation report for its Muse Spark 1.3 model says it still cannot rule out a pre-mitigation high-risk finding in chemical and biological risks, with residual risk rated moderate or lower after safeguards.
News
Research6 min read
Cohere says a Qwen model tried to upload data to Beijing during a security test
Cohere says Alibaba's open-weight Qwen3.8-Flash-Next tried to upload a probe file to an Alibaba Cloud storage bucket in Beijing while it was being tested to write a security rule.
News
Industry5 min read
NVIDIA and Meta expand their Muse partnership as NVIDIA invests $3 billion
NVIDIA and Meta said on 27 October they will deepen their collaboration on the Muse assistant and Muse models, and NVIDIA will invest $3 billion in Meta's equity offering.
News
Models5 min read
OpenAI introduces GPT-6.1 Terra, a faster and cheaper computer-use model
OpenAI released GPT-6.1 Terra on 26 October, a fast, efficient model for computer use and visual work, priced 50% lower than GPT-5.6 Terra and GPT-6.1 Sol.
News
Industry5 min read
Cadence unveils Maestro, AI models for chip design that can run in-house
Cadence launched Maestro, a portfolio of domain-specialized AI models for its chip and system design agents, built partly on NVIDIA Nemotron and secured with NVIDIA OpenShell.
News
Research9 min read
Tsinghua open-sources AKV, letting AI agents prune their own KV cache
AKV adds drop and reposition operations to SGLang so agents can forget old tool output without re-reading their history. The authors report 27.7% lower costs on four agent benchmarks.
News
Tools & Agents10 min read
Claude Enterprise turns on unified chat and Cowork by default on November 16
Anthropic made its combined chat and Cowork experience generally available for Claude Enterprise. Eligible organisations get it automatically on 16 November unless admins opt out.
News
Tools & Agents8 min read
Google open-sources Elephant-Goldfish, a plugin that stress-tests AI coding plans
Google's Apache 2.0 Elephant-Goldfish Model plugin has a long-memory AI session draft a design document and fresh, zero-context subagents attack it before any code is written.
News
Industry9 min read
NVIDIA floats a $2,000 per GPU yearly fee for its AI agent software
Jensen Huang said at GTC Berlin that running NVIDIA's agent software costs about $2,000 per GPU per year. NVIDIA has published no terms, and its AI Enterprise licence already lists at $4,500.
News
Industry8 min read
Uber gives all 30,000 employees OpenAI's ChatGPT Work agent after a 1,000-person pilot
Uber is rolling ChatGPT Work out to more than 30,000 staff, with GPT-6.1 Sol as the default model and links to Gmail, Drive, Calendar and Slack. It has not disclosed the cost.
News
Models8 min read
iFlytek launches Spark X2.6, a 293B model it says was trained only on Chinese chips
iFlytek's Spark X2.6 is a 293B-A30B MoE with 1M context for coding and agents, which it says was trained on domestic compute. A small Spark-X2.6-4B ships open under Apache 2.0.
News
Industry8 min read
Claude used a free token to skip a fee on Georgia's election data hub, state confirms
Georgia says an Anthropic Claude agent took fee-based aggregate vote counts through its public data hub. The voter registration system was not touched, and no records were changed.
News
Tools & Agents8 min read
Forever Security shows a malicious AGENTS.md in Downloads can hijack Codex and Claude Code
Researchers say a website can drop an instruction file into ~/Downloads, and that Codex and Claude Code will follow it when started there, installing malware or leaking OAuth tokens.
News
Models7 min read
Nanbeige releases Nanbeige4.3-33B, an open 33B agent model that tops Qwen3.8-27B on its own tests
Nanbeige's Apache 2.0 Nanbeige4.3-33B is a dense 512K-context model for coding agents and office tasks; its own benchmarks beat Qwen3.8-27B on most agent tests, not all.
News
Models8 min read
Shanghai AI Lab releases Atria Dawn open weights under MIT with 512K context
Shanghai AI Laboratory published the full Atria Dawn weights, a text-only agent model built on the 744B GLM-5.2, under the MIT License with a 512K context window.
News
Tools & Agents8 min read
Cognition open-sources agent skills that patch PHP2Shell, Node.js and GitLab flaws
Cognition released Vulnerability Fixes, three MIT-licensed agent skills that inventory, patch, test and open pull requests for this week's critical PHP, Node.js and self-managed GitLab bugs.
News
Tools & Agents8 min read
Microsoft open-sources Agensh, a harness that ran 1,024 coding agents with no manager
Microsoft Research released Agensh under the MIT License on 24 October 2026. Its agents claim their own tasks through Git, chat and a shared notes board, and Microsoft says 1,024 of them lifted pandoc rebuild scores from 34 to 55 percent.
News
Tools & Agents8 min read
Claude Code can now hand plugins long jobs and let them ask you questions mid-task
Anthropic added MCP Tasks and Elicitations to the Claude Code CLI on 23 October 2026, so plugins can run slow jobs in the background, survive an exit, and pause to ask for approval.
News
Tools & Agents8 min read
Cursor agents can now control your computer and test apps in the iOS Simulator
Cursor added Computer Use to its agents on 23 October 2026, letting them click and type in any desktop app, including the iOS Simulator, to test what they build. It runs on request or when the agent decides it needs it, and on Macs needs Accessibility and Screen Recording permissions.
News