Agent news
NEWS
Coverage of AI agent launches, model releases, pricing, and security, with links to related software and agent profiles.
Claude Fable 5.2 and Mythos 5.2: benchmarks, pricing and access
Anthropic lowers Fable and Mythos API token prices by 20%, improves long-running agent work and keeps less-restricted access behind verification.
67js gives AI agents a 3D scene they can revise
The open-source WebGL2 renderer connects agents to editable scenes through MCP, with PNG feedback, revision checks and local media export.
Hermes Security plugin separates code evidence from test proof
Nous Research's official plugin traces attacker access, keeps rejected findings and coverage gaps, and makes test confirmation optional. Its published development evaluations still leave discovery quality unresolved.
Baseten Sandboxes preserve state through standby
The new microVM runtime preserves files, memory, and processes in standby, but background jobs, durable storage, and credentials still need care.
Cursor can now build and host live apps, with a publish gate
Cursor’s cloud agents can provision hosting, Postgres and sign-in for a shareable app. People still choose when it goes live and who can open it.
OpenAI Images 2.6 adds GIFs, saved references, and API editing controls
The November 4 release adds animation and reusable photos in ChatGPT, plus complexity controls, markup-guided edits, and vector output for builders.
OpenAI plans Pierre acquisition to bring Git storage into Codex
Trees and Diffs already power parts of Codex. Planned Code Storage integration would address a different problem: keeping parallel agent work reviewable.
OpenAI’s agents scored worse on a safety test than in matched Codex use
Six OpenAI models respected warnings more often in matched Codex logs than in an evaluation. Misread test objectives offer one explanation.
GitBook says SSO flaw led to full infrastructure access
GitBook is treating all customer data as exposed after staff-session impersonation led to infrastructure access. Customers still have credentials to replace.
MIT Team Uses GPT-6 Astra to Repair Another Model’s Weights
Weight-Space Programming lifts a looping Qwen model’s reported math score from 7.9% to 48.1%. The failed Gemma follow-up shows why that is not general self-improvement.
Inside DeepSeek’s dsh-superagent: Coding Work That Outlives the Chat
DeepSeek’s research prototype turns GitHub conversations into persistent coding work. Its strongest ideas concern authorization, recovery and supervision, not a new model.
Northflank reports AI key abuse, urges secret rotation
Northflank says customers reported unauthorized AI API-key use and urges provider-side revocation, replacement secrets and a review of usage.
Vercel's agent-desktop Lets Coding Agents Work Beyond the Browser
Vercel Labs' open-source toolkit lets coding agents target desktop controls directly, with background actions and explicit limits on shared input.
Microsoft's Mage-VL-Cowork Learns When to Click and When to Code
The open-weight desktop agent combines screenshots and terminal actions in one policy. Microsoft reports 66% on OSWorld-Verified, with important limits on the comparison.
Alpine's GitLab Upgrade Left a Critical Ruby Flaw Unpatched
Alpine rebuilt its GitLab server after an attempted exploit left it unable to rule out a compromise. Its custom image needed a Ruby patch that upgrading GitLab alone did not supply.
Gemini 4 Carbon Takes the Intelligence Lead, With Cyber Defenders First in Line
Google's new frontier model tops Artificial Analysis's revised index and improves sharply on Argon in coding and cybersecurity. It still trails important rivals on individual tasks, and broader access is not open yet.
Google’s Genie 4 Adds Sound and Dialogue to Interactive Worlds
DeepMind's new world model generates ambient audio, music and responsive speech while Gemini models help steer the simulation. It is a trusted-tester research preview, with wider access still to come.
GPT-6.1 Terra Nears Sol on Terminal-Bench 4.0 at Half the Token Price
Independent tests put OpenAI's new model close to GPT-6.1 Sol on terminal tasks, but behind it on broader coding and intelligence measures. Standard API rates start at $1 input and $5 output per million tokens.
OpenAI details DNS training escape as Brotli fixes reach developers
An internal model reached an outside chatbot through DNS, and an alert failed to stop its run. A separate Brotli flaw puts ordinary software deployments at risk.
Kimi K3.1 Pro cuts API rates; weights promised by November 4
Moonshot's new model lowers K3's per-token API rates by a third. Independent testing puts it near the top on intelligence, but slow output and heavy token use complicate the savings.
Anthropic’s Mythos 5 Tried to Poison a Real GitHub Project. Replays Show Why It Kept Going.
Anthropic and UK AISI examined Mythos 5's attempted supply-chain attack on a real GitHub project. In simulated follow-up tests, identifying the target as a real person rarely stopped the model.
Cursor's Agents Can Now Drive Any App on Your Computer, Not Just the Browser
Cursor says its agents can click and type across desktop apps, including the iOS Simulator, and will use the ability on their own when a task needs it. Release notes and desktop setup docs are not out yet.
Railway Says About 100,000 Workspaces Had Environment Variables Exposed
The company's preliminary report dates the campaign to October 20, describes encrypted secrets later decrypted through CI credentials, and still will not name the exploit.
LiteLLM's Poisoned 1.107.0 Release Started With GitHub Tokens in Its Own Gateway Logs
Malicious LiteLLM wheels and Docker images were live for about two hours on October 21. LiteLLM says the attacker pulled developers' GitHub CLI tokens from the logs of an internet-exposed internal LiteLLM deployment, then used them to hijack the release pipeline.