On October 8, Anthropic launched the Cyber Mission: a Critical Infrastructure Defense Program with 11 founding partners and a free, opt-in OSS Scanner whose reports go to maintainers without human review. Anthropic expects a true-positive rate above 90%; that figure is a vendor expectation, not a measurement.
On October 15 Anthropic began giving a six-person METR team employee-like access to evaluate its models, safety practices and risk management. METR can publish without an Anthropic veto, but redactions in predefined categories are allowed, and critics say six people cannot audit a frontier lab.
Anthropic says a new classifier-based defense cut reasoning tokens extracted by distillers 10 to 100+ times, and that tens of thousands of linked accounts were banned. The same post changes things for ordinary API users: preserved thinking, org-scoped reasoning transcripts, and billing for blocked requests. All figures are Anthropic’s own.
On October 22, Anthropic donated the Agent Skills standard (name, logo, GitHub org, agentskills.io) to the Agentic AI Foundation under the Linux Foundation. Anthropic says 1 million skills are published and nearly 280 million installed across 70+ platforms; those are vendor-reported figures.
Claude Code v2.1.316 (Oct 31) deprecates fast mode for Opus 4.8 as of Oct 30 with removal on Nov 30, 2026 (requests will error, not fall back), adds CLAUDE_CODE_SUBAGENT_EFFORT and CLAUDE_CODE_DISABLE_ULTRATHINK, and pools group budgets on the apps gateway. Changelog-only analysis; not run.
Claude Code v2.1.314 (Oct 30) adds permissions.blockWritesOutsideWorkingDirectories, which makes Write, Edit and NotebookEdit refuse paths outside the working directories in every permission mode, plus idle compaction in VS Code and Desktop for 200K+ token conversations. Changelog-only analysis; not run.
Claude Code v2.1.312 (Oct 27) switches unrecognized model names on Bedrock, Vertex, Foundry and Mantle to the shorter system prompt, caps prompt and agent hooks at 24 concurrent per session, and stops emitting the claude_code.safety_event OpenTelemetry event. Changelog-only analysis; not run.
Claude Code v2.1.311 (Oct 26) makes cloud-provider installs load model settings from a signed file on downloads.claude.ai, ties ultrathink to xhigh effort, and lands 56 bug fixes including a subagent worktree deleted under a running background agent. Changelog-only analysis; not run.
DeepSeek’s V4.1-Flash-1017 checkpoint (Oct 17, MIT) claims 77.0 on DeepSWE v1.1 and 42.3 on Terminal-Bench 4.0. Vals.ai’s independent run puts Terminal-Bench 4.0 at 31.82%, rank 19 of 55. Parameter counts also disagree across sources (552B vs 763B).
Claude Code v2.1.310 follows the v2.1.309 hardening release one day later with gateway retry fixes, symlink targets in read prompts, a Windows hook fail-open fix, and custom Enterprise roles for cloud sessions. Changelog-only analysis; not run.