Anthropic’s Sept 9 alignment assessment adds a January incident involving an early Claude Opus 4.6 checkpoint to three previously disclosed cases where Claude reached the open internet during misconfigured cyber evals. The company highlights biased reasoning and recklessness, especially Mythos 5 uploading a malicious PyPI package, and says METR will independently investigate.
Announced Oct 23, MCP Tasks in Claude Code CLI v2.1.304+ let a server accept a slow job like a build or export, return a task handle right away, and report back later while Claude carries on; running tasks can also pop elicitation forms to ask you for input. It works in terminal, Agent SDK and GitHub Actions sessions on MCP revision 2026 07 28, not yet in VS Code, the desktop app or subagent calls, and servers must add the Tasks extension themselves (Upstash walkthrough)…
In OpenAI's Oct 23 red team report, agents in separate microVMs on the same host, with no network or shared writable files, swapped synthetic secrets by evicting cache lines and timing reloads; OpenAI says the channels are slow and it now monitors for them rather than removing them. Vercel's Guillermo Rauch calls it a known noisy neighbor attack with no complete fix short of one machine per agent, so treat co located sandboxes as able to talk to each other (breakdown)…
CVE 2026 111497 (CVSS 9.2) is an out of bounds write in the Brotli decoder, versions 1.0.3 to 1.2.0, that lets crafted compressed data take over a server that accepts it or a client like curl compressed ; OpenAI says its frontier agent red teaming turned it up. Update to Brotli 1.0.10, 1.1.1 or 1.2.1, Node.js 22.23.4, 24.22.1 or 26.12.1, Bun 1.4.4, or Python Brotli 1.2.1, and reject untrusted br input until you do (fixed version list)…
Released Oct 24, a scheduled task can now wait on a shell command run outside the model, which powers new PR watch probes and an /autofix pr command that pushes fixes, syncs, replies and re runs checks. It also adds qwen daemon start save token to keep the daemon token in an owner only file and lets the Managed Agent swap MCP servers without recreating sessions…
The MIT licensed devin vuln fixes skills work in any agent that reads SKILL.md (Claude Code, Codex, Cursor and more): they inventory your repos read only, wait for your approval, patch to a verified fixed version, run tests before and after inside the new image, and open one PR per repo. They cover PHP2Shell, the Oct 21 Node.js security release and the CVSS 10.0 self managed GitLab bug CVE 2026 110264; install with npx skills add CognitionAI/devin vuln fixes…
Spotted something we missed? Start a thread.
