Bitcoin Core Changes Security Rules as AI Makes Finding Code Bugs Dangerously Easy

Developers merged a policy change that ends the months-long embargo for flaws a chatbot could rediscover, after a summer in which AI found hundreds of bitcoin bugs before the humans did.

Bitcoin Core, the foundational software running the vast majority of the global bitcoin network, has fundamentally overhauled its vulnerability disclosure policy. Moving forward, the project will stop sitting on bug fixes for months when those flaws are simple enough for artificial intelligence to rediscover independently. The decision comes after maintainers merged a crucial change to the project’s security policy, marking a major turning point in how open-source cryptocurrency infrastructure adapts to the rapid advancement of automated code analysis.

Historically, Bitcoin Core operated under a rigid, tiered vulnerability grading system designed for an era when software security was strictly a human endeavor. Every reported bug was evaluated for its severity, dictating a lengthy timeline for when the patch would be made public. Low-severity flaws were typically revealed two weeks after the next major software version shipped to the public. Meanwhile, medium- and high-severity vulnerabilities remained strictly under embargo—hidden from the broader public and even many developers—until two weeks after the last affected, vulnerable version reached its official end-of-life status.

The foundational logic behind this long-standing security model was straightforward: shipping a fix quietly and delaying any detailed explanation prevented malicious actors from learning about a security hole before the individuals and entities running nodes—the computers responsible for checking and relaying every transaction on the bitcoin network—had sufficient time to upgrade their software. In theory, this "security through obscurity" approach bought the ecosystem vital breathing room to patch vulnerabilities safely out of the public eye.

However, that traditional rulebook was built for a world where finding subtle cryptographic and implementation bugs required specialized human skill, deep domain expertise, and extensive manual code review. The explosive proliferation and capability of modern artificial intelligence tools have systematically broken how that disclosure process works.

"We recently experienced an increase in the number of reports for security vulnerabilities that can be easily (re-)discovered," explained Antoine Poinsot, a developer at Chaincode Labs who authored the policy change proposal, in his official submission. "This is largely due to LLMs drastically reducing the bar to find certain classes of issues or identify covert fixes."

How the New Policy Works

Under the newly adopted framework, Bitcoin Core maintainers can formally flag a vulnerability as easy to discover during the initial triage phase. In practical terms, this classification applies to any security flaw that "can be found with a simple prompt to an LLM."

When such a bug is identified, the standard multi-month embargo is bypassed entirely. Instead, the fix is immediately backported to every software version that Bitcoin Core still actively supports. Developers then package the solution into an expedited point release, deploying the patch "as soon as possible" accompanied by a security advisory containing the necessary technical details.

Supporters of the change argue that the old methodology no longer serves its original purpose in the age of generative AI. "For these issues that are easy to find, this is the wrong tradeoff," the updated policy document states, explicitly evaluating the flaws of the previous embargo period. "The embargo period achieves little but it delays the availability of a fixed release."

At the same time, accelerating the release cycle introduces new engineering compromises. As Poinsot pointed out in his proposal, these rapid fixes "won’t have as much ‘bake time’ in testing before they reach users," meaning that patches will spend less time undergoing rigorous peer review and test-net deployment before being pushed out to live environments.

Not every developer in the decentralized community reached an immediate consensus on the exact boundaries of the new policy. Antoine Riard, a prominent and long-standing bitcoin security researcher, advocated for a more flexible approach. Riard argued that Core should retain the option of shipping a critical fix without an accompanying advisory, holding back granular technical details on a case-by-case basis to prevent immediate weaponization by bad actors.

Poinsot firmly rejected that compromise, highlighting the unintended dangers of keeping users in the dark. "Alerting users of a security vulnerability without any details could create more risk unnecessarily," he wrote, pointing specifically to the danger of "a rushed migration by users who would otherwise not be affected" by the underlying issue.

The Summer of AI Pressure

The urgency behind the policy shift is far from theoretical. It follows a tumultuous summer that exposed deep vulnerabilities across the broader cryptocurrency infrastructure, driven in large part by automated tooling.

In late July, a volunteer collective of 16 developers known as the Bitcoin Red Team deployed artificial intelligence models across 390 distinct bitcoin-related software projects. In roughly a day of automated scanning, the group logged an astonishing 4,962 findings, of which 85 were classified as critical security vulnerabilities.

The real-world fallout was swift and severe. Within days of the Red Team’s automated sweeps, a flaw in BTCPay Server that the group had previously reported through private channels began being actively exploited on live merchant servers. The ensuing attacks successfully drained the Lightning network nodes that various merchants relied on to accept small, instantaneous bitcoin payments.

The threat prompted emergency defensive measures. Core Lightning, one of the primary implementations of the Lightning network infrastructure, was forced to issue an urgent directive telling operators to immediately disconnect their nodes after AI-generated vulnerability reports submitted to its small development team turned out to be authentic.

Even larger financial losses struck the ecosystem from vulnerabilities that did not strictly require AI to uncover, though automated tools compounded the security crisis. A legacy bug in Coldcard hardware-wallet firmware, which had remained dormant since 2021, allowed attackers to systematically drain 594 bitcoin in just a 25-minute window on July 30. Total financial losses quickly escalated, nearing $114 million across four separate malicious sweeps. When Coldcard ultimately shipped its emergency firmware fix, the company publicly noted that AI-driven analysis had helped them catch additional secondary bugs along the way.

While no specific vulnerability has been officially flagged under Bitcoin Core’s brand-new policy yet, the reality facing developers is stark. In an ecosystem where any modern AI model can potentially regenerate a software secret or spot a latent code flaw in a single prompt, hiding known vulnerabilities behind months-long embargoes was a strategy with an expiration date.