On October 8, Anthropic launched the Anthropic Cyber Mission, a long-term framework aimed at defenders. It ships with two concrete programs: the Critical Infrastructure Defense Program for operational technology, and OSS Scanner for open-source projects. Two days after Project Glasswing was folded into the Cyber Verification Program.
OSS Scanner forwards whatever the model finds
Open-source projects can opt in and then receive periodic scans from Anthropic’s most capable models, free of charge. Each report carries a proof of concept for the bug, an explanation, and a suggested fix where one is available.
The decisive sentence is in the fine print: the reports are model-generated and sent without human review. Maintainers get them faster, and some will be wrong, a severity rating for instance. Anthropic expects a true-positive rate above 90 percent and says it will work on improving it.
The request came from maintainers themselves. Under Glasswing, Anthropic scanned hundreds of widely used projects, had humans triage the findings, then disclosed privately. Some maintainers with the capacity to triage at scale asked for everything the models had found, reviewed or not. For everyone else, human-verified disclosure stays.
The explicit model is Google’s OSS-Fuzz, which lands on a sore spot. Four days ago Google froze its own OSS reward program because of too many automated submissions. Anthropic is now adding more automated submissions, though to projects that asked for them.
The money goes to the places that pool the reports
Anthropic names what it funded: the Python Software Foundation, Alpha-Omega and OpenSSF through the Linux Foundation, and the Apache Software Foundation. Plus Akrites and Gold Eagle, which collect and coordinate vulnerability reports from many sources so maintainers don’t get buried. It comes out of the Defender Advantage Fund, launched in August, which also keeps OSS Scanner free.
Maintainers can separately apply to Claude for Open Source, which hands out free Claude Max subscriptions.
Critical infrastructure: eleven partners, no operators
The Critical Infrastructure Defense Program brings frontier models, on-site Anthropic engineers and in-house threat research to the providers operators rely on. Founding partners are Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Hitachi, Insane Cyber, Nozomi Networks, Palo Alto Networks, PwC and Rockwell Automation.
The selection is the argument: consultancies that run security programs, security companies that guard the networks, and manufacturers that build and patch the equipment. Operators themselves aren’t on the list. They arrive through their providers.
Anthropic explains the need at the level of the hardware. Controllers, control software and industrial networks are built to last decades and often can’t be taken offline to patch, so known vulnerabilities sit open for years. In rare cases, Anthropic writes, a fix may have to wait decades until it can be applied safely to running machinery.
Anthropic says itself that Glasswing wasn’t enough
The most striking line in the announcement is an admission. Under Glasswing, partners uncovered many vulnerabilities but hadn’t yet achieved “a sufficient reduction in cyber risk.” Finding got cheap. Verifying, prioritizing and fixing did not. Months often passed between a bug being found and being fixed.
Anthropic draws two uncomfortable conclusions from that. First, more findings only help where somebody can process them, hence the focus on the coordination bodies and on automated triage. Second, its own forecast that AI will favor defense in two years does not hold in the near term. The program sits in exactly that gap.