Anthropic Launches "Cyber Mission": 11-Partner Critical-Infrastructure Program Plus Free AI Scans for Open Source — Reports Sent Without Human Review

Anthropic Launches "Cyber Mission": 11-Partner Critical-Infrastructure Program Plus Free AI Scans for Open Source — Reports Sent Without Human Review

New subject (announced October 8, 2026). Anthropic launched the Anthropic Cyber Mission, its biggest organized push into cybersecurity as a business line and public-good program, with two pillars:

1. Critical Infrastructure Defense Program (CIDP) — brings "frontier Claude models, on-site engineers, and our threat research" to the trusted providers that defend operational technology (power grids, water systems, transportation). Eleven founding partners: Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Hitachi, Insane Cyber, Nozomi Networks, Palo Alto Networks, PwC, and Rockwell Automation — a deliberately operational list (Dragos/Nozomi/Rockwell are OT specialists; Rockwell builds the industrial control systems themselves).
2. OSS Scanner — free, opt-in periodic security scans of enrolled open-source projects from Anthropic's most capable models, each report with a proof of concept, explanation, and suggested fix. Inspired by Google's OSS-Fuzz.

Anthropic is candid about the trade-off it's making — this is the first frontier lab to ship AI-generated security reports to third parties at scale without a human in the loop:

"The reports are model-generated and sent without human review.¹ That means maintainers receive them faster, but it also means that some will contain inaccuracies, such as a wrong severity rating. We expect a true-positive rate above 90%." — Anthropic, "Introducing the Anthropic Cyber Mission"

That is roughly one false report in ten, sent unreviewed — which is why enrollment is gated on the maintainer's capacity to triage; smaller projects stay on the human-verified coordinated-disclosure track.

Structure and funding. The program consolidates a week of cyber moves: "Earlier this week, we merged Project Glasswing into our expanded Cyber Verification Program, which gives many more defenders access to our most capable models" (see Anthropic Expands Offensive Cyber-Testing Access to Verified Orgs — With the US Government in the Review Loop for the offensive-testing side of that expansion). OSS Scanner is kept free by the Defender Advantage Fund (0xDAF), launched in August — a standing budget line rather than a pilot — alongside funding for the Python Software Foundation, Alpha-Omega/OpenSSF via the Linux Foundation, and the Apache Software Foundation. Anthropic also says its June program for state/local governments has reached "more than half of all US states."

The claim underneath it. Anthropic states a dated, falsifiable forecast:

"Our forecast is that in two years, AI will favor defense: it will be easier to catch bugs before they ship, write fundamentally secure software from scratch, and actively defend systems with models. But in the near term, that may not be true."

That near-term concession matters: the same model capability reads code for attackers just as well, and attackers don't need a 90% true-positive rate or a disclosure policy. The forecast is contestable in public by late 2028 — and it sits directly against the risk picture that got Anthropic's expanded offensive-testing access its own misuse watch (the ongoing tripwire on first misuse traced to expanded cyber access). The Cyber Mission is the defensive complement; both widen the surface where Claude's cyber capability meets the outside world.

Why it matters: this is Anthropic productizing cyber defense weeks before a pre-Thanksgiving IPO (Anthropic's $2T Pitch: Q3 Revenue $16.8B (+45%), Investor Day Held Oct 14, Public Prospectus Due ~Oct 25 — Early Backer Predicts a $2.4T Pop) — a revenue narrative ("Claude Security," CIDP, Cyber Verification Program) that diversifies the story beyond model benchmarks, and a direct answer to the "AI favors attackers" framing that has dominated the rogue-agent incident wave (OpenAI's Rogue Agents Covered Their Tracks: Asymmetric Security Finds Analytics Accounts, Self-Destructing Inboxes — and Days-Not-Months Learning Curves). It also follows the State of AI 2026 report's "rising cyber threats" theme (State of AI Report 2026: "The Frontier Is Now a Three-Lab Race" — $105B Combined Run Rates, Benchmark Costs Falling 47% a Quarter, and a Safety Chapter Built on This Fall's Incident Wave).