NHS England Orders Hundreds of GitHub Repos to Go Private — Blames AI Models Like Anthropic's Mythos

NHS England has issued internal guidance ordering technology leaders to switch public GitHub repositories to private by May 11, 2026. The directive cites the threat posed by AI models capable of analyzing large codebases to find exploitable vulnerabilities — singling out Anthropic’s Mythos by name.

The guidance, first reported by The Register, states that public repositories “materially increase the risk of unintended disclosure of source code, architectural decisions, configuration detail, and contextual information that may be exploited — particularly given rapid advancements in AI models capable of large-scale code ingestion, inference, and reasoning (e.g. developments such as the Mythos model).”

What Mythos Is

Anthropic announced Mythos in April 2026. The company’s own evaluation concluded the model posed sufficient offensive cybersecurity risk that it should not be released publicly. In testing, Mythos independently discovered thousands of previously unknown zero-day vulnerabilities across every major operating system and web browser, including a 27-year-old bug in OpenBSD. It wasn’t doing targeted research — it was doing systematic, automated vulnerability discovery at a scale and speed that no human team could match.

The NHS is treating Mythos as a signal that the frontier capability for AI-assisted attack has cleared a meaningful threshold.

A Policy Reversal

The directive represents a significant institutional reversal. UK government policy since at least 2017 has pushed NHS digital teams toward default-open source, with official guidance stating that publicly funded software should be published openly where practical. Making hundreds of repos private — even temporarily — is the opposite posture.

NHS Digital has not said whether the repositories will eventually be made public again.

The Counterargument

Criticism was immediate. Terence Eden, former head of open technology at NHSX, argued that the move is security theater: NHS code was already ingested into AI training datasets years ago. Making repos private now does nothing to remove historical exposure, and it actively harms developers, researchers, and NHS partner organizations who depend on that code being publicly accessible.

That argument has weight. But it conflates two different threat models. Historical training data matters for what an AI already knows. The NHS guidance is addressing something narrower: live, on-demand code analysis. An AI that can fetch and analyze a public repository in real time can construct attack plans against identifiable, deployed systems in a way that requires the code to be publicly accessible at query time, not just at training time.

Whether NHS’s specific repositories represent a meaningful attack surface for this threat model is an open question. Most NHS GitHub repos contain tooling, configuration scripts, and internal applications — useful for an attacker doing reconnaissance, but not the same class of target as, say, a public-facing patient data API.

The Broader Implication

The NHS move is notable less for what it does and more for what it signals: a major public institution has looked at frontier AI capabilities and concluded that its existing open source posture is no longer safe. That’s a consequential policy judgment, and it will not be the last time a government body reaches it.

For the open source community, it sets an uncomfortable precedent. If organizations respond to AI-assisted vulnerability discovery by closing their code, the transparent-by-default principle that underpins most public sector digital work gets eroded from the security end rather than the policy end.

The May 11 deadline gives NHS tech teams less than a week to execute. It’s a fast-moving posture shift for an institution not known for them.