Anthropic AI Updates: October 9, 2026

1. Anthropic Launches the Cyber Mission With a Critical Infrastructure Defense Program

Anthropic. Anthropic announced the Anthropic Cyber Mission, a long-term effort to help defenders secure critical systems, starting with critical infrastructure and open-source software. Its first piece, the Critical Infrastructure Defense Program (CIDP), targets operational technology in power grids, water systems, and transportation networks, where systems often cannot be taken offline for patching. Instead of selling directly to operators, Anthropic delivers frontier Claude models, on-site engineers, and threat research through 11 founding security and industrial partners: Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Hitachi, Insane Cyber, Nozomi Networks, Palo Alto Networks, PwC, and Rockwell Automation. Anthropic also funded the Python Software Foundation, Alpha-Omega and OpenSSF (via the Linux Foundation), and the Apache Software Foundation. Source

2. OSS Scanner Offers Free, Model-Generated Vulnerability Scans to Open-Source Maintainers

Anthropic. Anthropic’s Frontier Red Team launched OSS Scanner, an opt-in service modeled on Google’s OSS-Fuzz that uses models including Claude Mythos instead of fuzzers to scan enrolled open-source projects periodically at no cost. Each report includes a self-contained reproducer, an explanation, a bisection to the introducing commit where possible, and a candidate patch when available. Reports are sent without human review. In validation, expert penetration testers reviewed 97 critical and high-severity findings across 48 projects and found that 85 (88%) met Anthropic’s disclosure bar, and wolfSSL reported that 72 of 74 reports it received were valid. Core maintainers of critical projects opt in by opening a pull request against the anthropics/oss-scanner GitHub repository. Source

3. Claude Adds Live Dashboards and Motion Explainers, Moves Docs, Slides, and Design Out of Beta

Anthropic. Claude Dashboards, in beta on paid plans, builds a dashboard from a plain-language question over sources such as Snowflake, BigQuery, Databricks, Redshift, ClickHouse, and Salesforce, keeps it refreshed, and shows the query behind any number. Claude Motion, in beta for Team and Enterprise and invoked with /motion, generates short code-based animations (animated charts, report summaries, product walkthroughs) that export as MP4, with no video generation model involved. Docs, Slides, and Design are now out of beta on all plans including Free, and Anthropic says over 45 million docs, decks, and designs have been created so far. Claude Design is moving inside Claude conversations, and the standalone claude.ai/design site shuts down on December 14. Source

4. Anthropic Commits $150 Million to the Genesis Mission

Anthropic. Anthropic committed $150 million over three years to the Genesis Mission, the federal initiative to accelerate scientific discovery with AI, extending Claude access to more than 15 participating agencies including NASA, NIH, and NSF. The commitment covers Claude, Claude Code, and API credits for several hundred Genesis Mission research projects, partnerships on fusion energy and quantum computing, and onboarding support for scientists. Anthropic announced it at the White House Office of Science and Technology Policy’s Science: A New Golden Age Summit. Source

5. 2026 Usage Policy Update Rewrites Surveillance, Elections, and Weapons Rules

Anthropic. Anthropic published an updated Usage Policy and Supported Regions Policy, effective November 12, 2026. Changes include a new section on deceptive campaigns and artificial activity, a narrower elections section that drops the blanket ban on personalized campaign targeting, a weapons rule that explicitly covers weapons software and arming drones, and a rewritten surveillance section that bars using Claude to decide who to investigate, arrest, or charge. New high-risk requirements apply when Claude controls hardware that takes autonomous physical actions: an operator must be able to observe and stop it, and it must hold a safe state if Claude disconnects. The policy also adds a narrow prohibition on sustained, purposeless abuse of the model. Source

6. Claude Science Agents Produce the First Complete Ultraviolet Map of the Sky

Anthropic. Anthropic researcher and Johns Hopkins astrophysicist Brice Ménard used Claude Science to build what Anthropic calls the first complete sky map in far-UV (154 nm) and near-UV (232 nm) light. Parallel Claude agents downloaded and cross-calibrated surveys from GALEX, Swift, FIMS/SPEAR, and TD-1, then inpainted the roughly one third of the sky never observed in UV by learning its relationship to visible, infrared, and radio data. In hold-out tests, predictions came within about 10% of real measurements. The map labels each pixel as measured or predicted, includes uncertainty estimates, and adds UV estimates for over 100 million stars derived from Gaia data. Source

7. Block Uses Claude Fable to Orchestrate About 1,000 Pull Requests per Migration

Anthropic. In a case study, Block’s Head of AI Capabilities Bradley Axen described using Claude Fable 5 for design work (data models, API specs) while it directs dozens of cheaper models such as Opus or Sonnet to edit files and run tests, merging about 1,000 PRs in a typical migration with no reported drop in quality. Block benchmarks every use case across models and effort levels, treating effort as a grid to optimize, and is building an auto-selector to route tasks to the right model. Merges to main and production deploys still require two human approvals, and Block reports that Claude refused to help bypass that dual-approval check in testing. Source

8. Claude Code 2.1.295 Adds Fail-Closed Hooks and Terminal Status Reporting

Anthropic. Claude Code 2.1.295, tagged on October 8, lets command and HTTP hooks set onFailure: "block", so a hook that fails to start, times out, or exits unexpectedly blocks the action instead of letting it through. It also supports the Program Status Protocol (OSC 7501), letting compatible terminals show whether Claude Code is working, waiting for input, or done. For mod authors, the release adds a $.ui.notify call and support for children in a mod’s Button, and the Claude apps gateway gains an upstream time-to-first-byte timeout and per-upstream model lists. Version 2.1.294, tagged the same day, fixes instruction-style prompt and agent hooks that could allow actions they were meant to block. Source