Anthropic Launches Cyber Mission With 11 Founding Partners
TL;DR
- The 11 founding partners span OT security (Dragos, Nozomi Networks), consulting (Accenture, Deloitte, Booz Allen, PwC), enterprise security (CrowdStrike, Palo Alto Networks), and hardware (Rockwell Automation, Hitachi), covering the full ICS vendor stack.
- Anthropic identified 29,000 candidate vulnerabilities in six months but manually reviewed only about 6,000, leaving the majority unprocessed and outside coordinated disclosure workflows.
- Of 97 high- and critical-severity findings reviewed across 48 projects, 85 cleared disclosure criteria, 11 were duplicates, and 1 was invalid, yielding roughly 87% precision.
Anthropic on October 8 launched what it is calling the Cyber Mission, which the company describes as "a long-term commitment to securing the systems everyone depends on." The announcement names eleven founding partners for a Critical Infrastructure Defense Program: Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Hitachi, Insane Cyber, Nozomi Networks, Palo Alto Networks, PwC, and Rockwell Automation. Participating providers get "frontier Claude models, on-site engineers, and our threat research" to defend operational technology running power grids, water utilities, factories, and transportation networks.
The pitch is a mismatch, stated flatly. "Highly cyber-capable AI models are widely available to attackers now. But defensive tools—including our own—have not yet reached enough of the defenders who need them." The post adds that "state-sponsored adversaries have spent years gaining footholds in these systems, across many sectors, so that they can disrupt them."
The second launch is an OSS Scanner. It is a free opt-in service that runs periodic scans on enrolled open-source projects and returns a bug report with a working exploit and a proposed patch. "Each report includes a proof of concept of how the bug could be exploited, an explanation, and a suggested fix where one is available." Anthropic says it expects a true-positive rate above 90%.
The caveat sits in the same section. "The reports are model-generated and sent without human review. That means maintainers receive them faster, but it also means that some will contain inaccuracies, such as a wrong severity rating." Core maintainers can enroll at red.anthropic.com/oss-scanner.
The post credits earlier work, including Project Glasswing and an expanded Cyber Verification Program, plus funding Anthropic has routed to the Python Software Foundation, Apache Software Foundation, and Linux Foundation entities. It is the company's second safety-oriented post on our Anthropic tracker today, after this morning's updated usage policy on Claude abuse and election deception.
What others are reporting
-
Axios Read →
Pre-briefing exclusive ahead of the official announcement, providing inside sourcing on the strategic framing and program structure before the public release.
-
Help Net Security Read →
Features direct user feedback from OpenSSL Corporation on scanner quality vs. human researchers, plus operational detail on enrollment criteria and the 85/97 disclosure approval rate.
Early AI reports...were as good and sometimes better than what we get from people. — Anton Arapov, OpenSSL Corporation
-
Cyber Security News Read →
Frames the mission around Project Glasswing's finding that detection speed is not the real bottleneck, and flags that removing human verification is a deliberate calculated risk.
The effort combines advanced Claude models, engineers, threat research, and funding, with an emphasis on fixing weaknesses
-
GBHackers Read →
Reports the 29,000 candidate vs. 6,000 reviewed vulnerability gap and the U.S. state government program reach, both absent from Anthropic's official release.
its government cyber defense program has supplied models and technical support to more than half of U.S. states
Originally reported by anthropic.com
Read the original article →Original headline: Anthropic Launches Cyber Mission With Critical Infrastructure Program and 11 Founding Partners