Bitcoin Core ends long embargoes for bugs that AI can easily rediscover

Bitcoin Core has revised its security disclosure policy to accelerate fixes and disclosures for vulnerabilities that large language models can easily rediscover. The change ends months-long embargoes for those vulnerabilities. Maintainers merged the revision on Thursday. Chaincode Labs developer Antoine Poinsot, who authored the proposal, announced the change that day through the Bitcoin Development Mailing List.

Poinsot said reports of easily rediscovered security vulnerabilities had increased recently. Poinsot attributed much of the increase to large language models making certain classes of vulnerabilities and concealed fixes much easier to identify.

Bitcoin Core runs on most nodes across the Bitcoin network. The previous policy linked disclosure timing to the severity assigned to each reported vulnerability. Low-severity vulnerabilities became public two weeks after the following major version was released. Medium- and high-severity vulnerabilities remained confidential until two weeks after the final affected release reached end of life, potentially extending the embargo to a year. Those timelines allowed node operators, many of whom upgrade slowly, to install patched software before public disclosure. In practice, fixes were not released for older maintained versions and instead appeared in the following major release. Poinsot argued that this compromise does not work for easily rediscovered vulnerabilities. The policy says an embargo offers little benefit for such vulnerabilities while postponing access to a fixed release.

Bitcoin Core can now classify a vulnerability as easily discoverable during triage. The category covers vulnerabilities requiring little effort to find, including through a straightforward prompt to a large language model. Vulnerabilities discovered through other methods can also qualify. Bitcoin Core will backport fixes for qualifying vulnerabilities to every maintained major version. Bitcoin Core will issue point releases as quickly as possible. Bitcoin Core will publish a security advisory containing some details when the patched software becomes available. Poinsot acknowledged that the accelerated fixes will undergo less testing before users receive the fixes.

Vulnerabilities that do not qualify retain the existing severity-based disclosure timetable. Critical vulnerabilities remain exempt from the standard policy and will receive individual treatment. Critical vulnerabilities include flaws that could permit protocol-level coin theft, increase the coin supply or permanently split the chain.

Developer Antoine Riard proposed allowing fixes without accompanying advisories and withholding details when individual circumstances warrant. Poinsot rejected the proposal. Poinsot argued that warnings without details could unnecessarily increase risk by encouraging unaffected users to migrate hastily.

The policy revision follows months of increased pressure on Bitcoin's open-source projects. In late July, the Bitcoin Red Team, a volunteer group comprising 16 developers, used AI models to examine 390 bitcoin repositories in roughly one day. The group recorded 4,962 findings, with 85 classified as critical. Within days, attackers exploited a BTCPay Server vulnerability that the group had privately reported, draining Lightning nodes on live merchant servers. In August, Core Lightning developers instructed operators to upgrade or disconnect after AI-generated vulnerability reports proved correct.

The Bitcoin Red Team's work followed an exploit involving Coldcard hardware wallets. Attackers drained 594 BTC within 25 minutes on July 30. Losses from four sweeps totaled nearly $114 million.