On October 8, 2026, Anthropic announced the Anthropic Cyber Mission, which it describes as a long-term commitment to securing the systems everyone depends on. The initiative aims to give defenders tools, research and resources to secure their software and systems, and it starts on two fronts: critical infrastructure, with a new Critical Infrastructure Defense Program (CIDP), and open-source software, with a service called OSS Scanner.
The company justifies the move with a blunt diagnosis. Frontier models can be misused to exploit vulnerabilities and conduct cyber operations, and many areas of technology remain dangerously exposed. According to Anthropic, state-sponsored adversaries have spent years gaining footholds in these systems so that they can disrupt them, while critical infrastructure defenders and the open-source community face severe resource shortages that this moment makes worse.
The announcement builds on the experience of Project Glasswing, the program in which Anthropic partners used its models to find flaws. The company admits that, although many vulnerabilities were uncovered, a sufficient reduction in cyber risk has not yet been achieved: finding flaws is easier than ever, but verifying, prioritizing and fixing them remains hard. Earlier that same week, Anthropic merged Glasswing into its expanded Cyber Verification Program, which gives many more defenders access to its most capable models.
The first front is so-called operational technology: the controllers, control software and industrial networks that run power grids, water plants, factories and transportation. These systems are built to last for decades and often cannot be taken offline to patch, so known vulnerabilities can stay unresolved for years. Securing them is specialized work: the equipment is proprietary, changes are risky and a mistake can take down a plant.
The CIDP brings frontier Claude models, on-site engineers and Anthropic’s threat research to the trusted providers that operators rely on. Its founding partners are Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Hitachi, Insane Cyber, Nozomi Networks, Palo Alto Networks, PwC and Rockwell Automation. The list combines consultancies that run security programs, companies specializing in protecting business and industrial networks, and manufacturers that build and patch the equipment itself.
Anthropic says the work is already under way and that several partners are using Claude to fix vulnerabilities and help their customers do the same. The company acknowledges that critical infrastructure is hard to defend for many reasons AI cannot fix, but argues that frontier models can help find and repair weaknesses before they are used to cut off power or make water unsafe. The first step will be to work with a small cohort of providers to learn which strategies work best.
Partners backed the program with statements. Sam Rubin of Palo Alto Networks’ Unit 42 said attackers are using AI to find and exploit exposures “at machine speed.” Dan Gunter, founder of Insane Cyber, noted that the sites that need protection most, such as remote substations, offshore platforms and air-gapped plants, are exactly where traditional tools cannot reach. Daniel Bernard of CrowdStrike summed up that machine-speed threats require machine-speed defense.
Anthropic also recalled that in June it launched a cyber defense program for U.S. state, local, tribal and territorial governments. Since then, it says, it has offered frontier Claude models and technical support to more than half of all U.S. states and to some of the country’s largest public critical infrastructure operators, speeding up tasks such as code scanning and patching, incident response and red teaming.
The second front is open source, on which almost all software depends and which is often maintained by small teams of volunteers. Under Glasswing, Anthropic scanned hundreds of widely used projects, had humans review many of the potential vulnerabilities and privately reported them to maintainers. Some maintainers with the capacity to triage at scale asked to receive everything the models had found, reviewed or not. OSS Scanner grew out of that request.
OSS Scanner is an opt-in service inspired by Google’s OSS-Fuzz. Enrolled projects receive periodic scans from Anthropic’s most capable models, including Claude Mythos, free of charge. Each report includes a proof of concept of how the bug could be exploited, an explanation and, where possible, a suggested fix. Reports are model-generated and sent without human review, which makes them faster but also prone to errors, such as a wrong severity rating. Anthropic expects a true-positive rate above 90%.
That is why the service is meant for projects with the capacity to keep up with findings; for others, the company will continue to send human-verified disclosures under its coordinated vulnerability disclosure policy. Core maintainers of eligible projects enroll by submitting a request to a public repository, and eligibility is decided case by case using criteria similar to OSS-Fuzz’s, focused on projects with a critical impact on infrastructure and user security.
Anthropic also detailed its financial support for the ecosystem: it funded the Python Software Foundation, Alpha-Omega and OpenSSF through the Linux Foundation, and the Apache Software Foundation, and it supports Akrites and Gold Eagle, which coordinate vulnerability reports so maintainers are not overwhelmed. The Defender Advantage Fund, launched in August, finances pilot programs and keeps OSS Scanner free. Maintainers can also apply for free Claude Max subscriptions.
The company offers a forecast that sums up its bet: in two years, AI will favor defense, because it will be easier to catch bugs before software ships, write secure code from scratch and defend systems with models. In the near term, however, that may not be true. The cost of exploiting vulnerabilities has dropped, while verifying, disclosing and fixing them is still slow and depends on people. In Glasswing, according to Anthropic, months often passed between discovery and fix, and in operational technology a patch can take years.
In short, Anthropic is trying to put the offensive and defensive capabilities of its models in the hands of those who protect electricity, water, transportation and shared code. Success, by the company’s own measure, will mean those services keep running in the face of capable attackers, with fewer ways in and faster recovery. Anthropic warns that the effort will take years, that it will also share what does not work, and that it expects other developers, security companies and governments to launch initiatives of their own.
