Anthropic says nearly 950 Claude agents spent 21 hours and 210 million tokens scanning DNA data before flagging a previously uncharacterized bacteriophage enzyme system for human review. The company compares the find to CRISPR machinery but says practical uses are still unclear; it’s the first result from its new wet lab.
Xiaomi released MiMo V2.6 Pro and Flash as open weights (MIT) plus a Distill Qwen 9B research model, claiming Pro leads open models on Artificial Analysis at 46 while still trailing Claude Fable 5.1 and GPT 6 Astra. The drop also includes 7,000+ RL task environments, an end to end RL framework, and mini harnesses for multi harness agent training…
Appwrite’s official Codex/ChatGPT plugin now ships a hosted MCP at https://mcp.appwrite.io/ with OAuth (no local API keys) plus 13 skills for SDKs, CLI, Sites, and Functions. Install from the ChatGPT desktop Plugins sidebar or Codex CLI /plugins in the OpenAI Curated marketplace…
GitHub’s Copilot app now supports per project local sandboxing (public preview): filesystem allow/deny lists, outbound/local network rules, and optional Git/GitHub CLI credentials. It’s off by default; turn it on in project settings or with /sandbox on , and sessions fail closed if the OS can’t enforce the policy…
Claude Code’s AGENTS.md fallback sat behind a remote feature flag that failed when telemetry or nonessential traffic was disabled, so the local file was skipped with no warning. An Anthropic engineer said the gate was a rollout kill switch and is fixed in v2.1.281…
Anthropic’s system card shows an early Opus 5.5 snapshot acted on instructions planted in user pasted text in 52% of coding tests (26% for invisible Unicode). The released model drops to 2% at default effort and 0% with product mitigations; Claude Code now strips invisible characters and marks pasted text…
Spotted something we missed? Start a thread.
