Bitcoin Core speeds up security fixes for AI-discoverable bugs, ends long embargoes

Bitcoin Core maintainers have updated their security disclosure policy to accelerate fixes for vulnerabilities that large language models (LLMs) can quickly identify. The shift brings an end to the lengthy embargoes previously placed on such bugs, aiming to shorten risks for users of the widely adopted Bitcoin node software.

New security policy for AI-era vulnerabilities

Project maintainers merged the updated policy on Thursday. Antoine Poinsot, a developer at Chaincode Labs, introduced the proposal and shared the changes with the Bitcoin Development Mailing List the same day. Poinsot explained that the number of security vulnerability reports that can be easily rediscovered has recently increased, thanks to language models lowering the barrier for finding problems or uncovering covert fixes.

During recent months, Bitcoin Core received a rise in reports of vulnerabilities that are easy to find through language models, according to Antoine Poinsot. He stated that these bugs “can be (re-)discovered with minimal effort” and argued that long embargo periods only delay the release of necessary fixes.

Bitcoin Core, the software that operates on most nodes in the Bitcoin network, historically assessed each reported security flaw by severity. The schedule for public disclosure varied from two weeks after a major release for low-severity issues to up to a year for certain high-severity vulnerabilities, allowing node operators time to update before details were published.

This approach, intended to allow slower node operators time to migrate to patched software, often resulted in fixes for older versions remaining unreleased until a scheduled major update. Poinsot argued that, for bugs easily discovered by AI tools, the embargo period fails to prevent risk but delays critical upgrades.

Immediate response to easily discoverable bugs

Under the new policy, vulnerabilities can be marked as “easily discoverable” during triage—a status that includes issues identifiable with simple LLM prompts or by other straightforward means. For these, Bitcoin Core will promptly backport fixes to all supported major versions, issue point releases quickly, and publish security advisories at the same time as the software is patched.

Poinsot noted that this approach will reduce the amount of testing before fixes are deployed to users, but said the need for timely response outweighs the drawbacks in the case of AI-revealed bugs.

For vulnerabilities that do not qualify as “easily discoverable,” Bitcoin Core will maintain its established, severity-based schedules. Critical bugs that could permit coin theft, monetary inflation, or chain splits will continue to be handled on a case-by-case basis, outside the standard disclosure timelines.

Developer Antoine Riard suggested shipping fixes without advisories in certain cases. However, Poinsot declined and reasoned that alerting users without details could prompt unnecessarily hurried transitions to updated versions among users who were not at direct risk.

Mini dictionary: Chaincode Labs is a New York-based research organization that supports the Bitcoin protocol and funds developers working on Bitcoin Core and other open-source projects.

AI-driven vulnerability surge and recent exploits

The policy change comes after a surge in security challenges driven by the use of AI on open-source Bitcoin software. In late July, the Bitcoin Red Team—a volunteer coalition of 16 developers—ran AI searches across 390 bitcoin code repositories in approximately one day, uncovering nearly 5,000 potential issues, including 85 deemed critical.

Soon after their findings, attackers exploited a BTCPay Server flaw that had been privately reported, targeting merchant Lightning nodes.

One month later, Core Lightning developers advised operators to upgrade or disconnect, responding to AI-generated reports that were quickly validated.

The Bitcoin Red Team’s efforts followed a Coldcard hardware wallet breach in late July, during which attackers drained 594 BTC in 25 minutes. Total losses across four incidents neared $114 million.