Anthropic AI Updates: October 16, 2026

1. Anthropic Gives METR Team Employee-Like Access for Independent Risk Assessments

Anthropic. Anthropic has onboarded six METR employees with company hardware, office space, and access to internal tools roughly on par with its own risk assessment teams, excluding narrow categories such as customer data, privileged legal information, and model architecture details. METR will publish its analysis of Anthropic’s risk levels, incidents, and practices, and Anthropic cannot veto publication, though it can require redactions in predefined categories that METR may publicly characterize. The company describes three tiers of evaluator access, with SecureBio reviewing the chemical and biological weapons chapter of its August Risk Report and Accenture and Faculty working toward red-teaming and alignment assessments. Anthropic says it is also in early discussions with other organizations, including Hugging Face. Source

2. Claude Code Review Exits Research Preview for Team and Enterprise Plans

Anthropic. Code Review is now generally available for Claude Team and Enterprise organizations, with most reviews finishing in about 6 to 10 minutes compared with roughly 20 minutes during the preview. Admins can set each repository to Quick, Standard (default), or Deep review levels, and findings are now written as short statements of the problem, its trigger, and the fix. Setup requires enabling Code Review in Claude Code settings, installing the GitHub App, and selecting repositories, after which reviews run automatically on new pull requests. Eligible Enterprise organizations can claim at least $10,000 in Code Review credits by January 15, 2027, which expire 30 days after claiming. Source

3. Claude Code 2.1.301 Exposes Auto Mode Classifier Decisions in OpenTelemetry

Anthropic. Claude Code 2.1.301 adds classifier_verdict, permission_mode, and (with OTEL_LOG_TOOL_DETAILS=1) classifier_category fields to OpenTelemetry tool_decision events, so auto mode classifier blocks can be distinguished from permission rules. A new CLAUDE_CODE_RETRY_WATCHDOG_MAX_USAGE_LIMIT_WAIT_MS setting lets unattended retry mode end a request when a usage limit resets too far in the future. The release also adds OpenTelemetry tracing guidance to the claude-api skill, the % bracket-matching motion in vim mode, multi-line rows input fields for mods, and LaTeX math rendering in the VS Code extension. Source