Anthropic is rolling out a new security push aimed squarely at governments that are short on time, staff and money but still expected to keep hackers out. The company has launched a program that lets state, local, tribal and territorial governments use its Claude models to hunt for and fix security flaws, backed by what it says is up to $15 million in usage credits. The pitch is simple: credits plus hands-on help so public-sector IT teams can find vulnerabilities before attackers do. California and Texas are already in the mix.
Under the plan, Anthropic will allocate up to $15 million in Claude credits and let smaller SLTT governments apply for awards of up to $100,000, with additional slots reserved for larger local entities that support critical infrastructure, as reported by GovTech. Participating agencies are set to join cohort-based engagements that feature weekly technical training and access to runbooks tailored for scanning, triage and remediation of whatever issues Claude turns up.
Michael Lai, Anthropic’s lead on AI for state and local government, told StateScoop that state and local governments run the elections, benefits systems, and emergency services Americans rely on every day, and said the program is designed to help jurisdictions find and fix vulnerabilities. The company also told reporters that the first cohort kicked off last Thursday, bringing in representatives from multiple state and local agencies.
How This Differs From Project Glasswing
Anthropic says the credits program operates separately from Project Glasswing - the tightly controlled effort that gives vetted partners access to the Mythos research preview - and instead relies on publicly available Claude products and defensive runbooks for bulk scanning. The company’s own Red Team post explains why Mythos was gated in the first place: the preview showed unusually strong autonomous vulnerability-discovery and exploit-generation capabilities, so Anthropic limited access while it coordinates defensive disclosure and remediation work, as outlined in Anthropic.
Who’s Already In
California and Texas are among the early participants, and state officials are framing the program as a chance to speed up the defensive side of the arms race. California state CISO Vitaliy Panych said in a statement that having industry and government working together to test and validate what’s possible is preparing for the future together, while Texas Cyber Command Chief TJ White warned that bad actors will try to operationalize AI-driven attacks, according to StateScoop.
What This Means For Cities And Counties
Local IT teams often juggle aging systems, thin staff and inconsistent funding for incident response. Anthropic’s package of credits, training and cohort support is pitched as a way to close some of that gap, according to reporting that reviewed the company’s outreach email and application survey. GovTech obtained the program email, which describes a $100,000 credit allocation per eligible SLTT entity and a capped cohort structure meant to help agencies scan, triage and patch systems using Claude Security, Claude Code and Opus models.
Anthropic and other industry players say the initiative is about speeding remediation and boosting defender capacity as AI accelerates vulnerability discovery. The company presents the SLTT program as a way to get its defensive workflows and runbooks into real government environments, in line with its Red Team discussion of responsible disclosure and triage processes. Anthropic stresses the need for coordinated disclosure and human triage when models surface large volumes of potential flaws.
