Bitcoin Core Changes Security Policy to Counter AI-Rediscovered Vulnerabilities

Bitcoin Core’s security approach has shifted to address an uptick in vulnerabilities that can be easily rediscovered using modern AI tools. The maintainers updated their disclosure policy so that certain issues identified as trivial for large language models (LLMs) will be fixed and released promptly across all supported versions, rather than being held in embargo.

Previously, Core assigned severity levels to reported vulnerabilities and delayed public disclosure to let node operators update first. That model assumed attackers needed time and skill to find flaws, but LLMs have shortened that discovery window for some classes of problems.

New triage criteria and faster fixes

Core contributors can now mark a report as easy to discover during triage — effectively, a bug that “can be found with a simple prompt to an LLM.” When an issue meets this threshold, its patch must be backported to every supported release and shipped in a point release “as soon as possible,” accompanied by a security advisory containing meaningful details.

The rationale is pragmatic: keeping fixes secret under embargo provides little added security for bugs any model can reproduce, while it delays users from receiving a patch. At the same time, maintainers acknowledge a tradeoff: fixes deployed faster will have less bake time in testing before reaching users, increasing the risk of regressions.

Debate among maintainers

Not all developers agreed on the degree of transparency. Some wanted to retain the option of shipping a fix quietly without publishing advisory details in every case. Others pushed back, arguing that notifying users of a vulnerability without sufficient context could provoke needless migrations or other risky behavior.

Antoine Poinsot, who proposed the policy update, argued against silent fixes: he warned that partial alerts could increase risk for users who might react rashly without understanding whether they were affected.

Context from recent incidents

A volunteer group calling itself the Bitcoin Red Team ran LLM-based scans across hundreds of bitcoin projects and produced thousands of findings in a short period; dozens were rated critical. Within days of disclosure, some reported issues began to be exploited in the wild, demonstrating how quickly automated checks can surface real problems.

Other incidents illustrate the stakes. A BTCPay Server flaw reported privately was exploited on merchant systems, draining funds from Lightning nodes. Separately, a long-dormant Coldcard firmware vulnerability allowed attackers to sweep large amounts of bitcoin within minutes; Coldcard’s subsequent fix credited AI-assisted discovery in its broader audit process.

What this means for users

So far, no vulnerability has been publicly classified under the new “easy to discover” rule. Still, the update recognizes an inevitable reality: secrets a model can reconstruct from a single prompt won’t stay secret for long. The policy favors quicker, wider distribution of fixes and clearer advisories for things that AI can easily re-find, while accepting some reduction in pre-release testing time.

Maintainers and operators will need to balance speed of deployment, transparency, and testing rigor as automated discovery tools continue to evolve.