Bitcoin Core overhauls security policy as AI makes bug discovery trivial

New rule ends months-long embargo for vulnerabilities an AI can easily rediscover, speeds patching but reduces testing time.

By LineZotpaper

Published23 October 2026

Read Time3 min

Bitcoin Core, the primary software powering the bitcoin network, will stop keeping security fixes under months-long embargo when the bug is one an artificial intelligence model could quickly find again. The policy change, merged on Thursday, replaces delayed disclosure with immediate patches and security advisories for vulnerabilities that can be "found with a simple prompt to an LLM." The decision follows a surge in AI-assisted vulnerability discovery, including an audit in July that flagged nearly 5,000 potential flaws across 390 bitcoin projects, 85 of them critical.

Previously, Bitcoin Core ranked every vulnerability by severity and timed disclosure accordingly. Low-severity flaws were revealed two weeks after the next major version shipped; medium and high ones stayed under embargo until two weeks after the last affected version reached end of life. The logic was to give node operators time to upgrade before attackers learned of the hole. With the spread of AI tools, maintainers concluded that long embargoes for easily found bugs no longer served their purpose.

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

Under the new rule, once a vulnerability is flagged as easy to discover (in practice, one that can be found with a simple prompt to a large language model), developers will copy the fix to every supported version and ship a point release "as soon as possible," accompanied by a security advisory with some details. The policy acknowledges that these fixes "won't have as much 'bake time'" in testing before reaching users.

Not every developer agreed with the approach. Antoine Riard, a longtime bitcoin security researcher, wanted the option to ship a fix without an advisory and hold details back on a case-by-case basis. Poinsot refused, writing that "alerting users of a security vulnerability without any details could create more risk unnecessarily" and might trigger a rushed migration by users not actually at risk.

The change comes after a summer of AI-driven security incidents. A volunteer group of 16 developers calling itself the Bitcoin Red Team ran AI models across 390 bitcoin projects in about a day in late July and logged 4,962 findings, 85 of them critical. Within days, a BTCPay Server flaw the group had reported privately was being exploited on live merchant servers, draining Lightning nodes used for small bitcoin payments. Core Lightning, another major implementation, told operators to disconnect because AI-generated reports overwhelmed its small team.

The most costly incident struck Coldcard, a hardware wallet, where a bug that had lain dormant since 2021 enabled thieves to drain 594 bitcoin, worth nearly $114 million, in 25 minutes across four sweeps on July 30. While that flaw required no AI to find, Coldcard said AI had helped it catch more bugs along the way.

No bug has yet been flagged under the new Bitcoin Core rule, but as Poinsot's proposal noted, a secret that any model can regenerate in a single prompt was never going to protect users for long.

Analysis

Why This Matters

The security of the Bitcoin network depends on node operators updating quickly. Frequent out-of-cycle patches may increase operational load but reduce the window for exploitation.
AI's ability to rediscover vulnerabilities is forcing open-source projects to rethink their disclosure practices. Bitcoin Core's move could set a precedent for critical infrastructure software.
With billions of dollars in cryptocurrency at stake, the tradeoff between secrecy and speed has real financial consequences for exchanges, merchants, and holders.

Background

Bitcoin Core is the reference client for the Bitcoin network, maintained by a community of volunteer developers. Its previous vulnerability disclosure process was built for a world in which finding a security flaw required significant expertise. The policy graded every bug by severity and scheduled public disclosure months or years after a patch shipped, aiming to give node operators time to upgrade before attackers could reverse-engineer the fix. The rise of large language models capable of scanning code and flagging anomalies with minimal human effort undermined that strategy. A coordinated AI audit in July 2026 of 390 bitcoin projects netted nearly 5,000 potential flaws, and real-world exploits followed within days, prompting the maintainers to conclude that long embargoes for easily found bugs were no longer effective or defensible.

Key Perspectives

Maintainers (led by Antoine Poinsot): The old embargo system is a wrong trade-off when a bug can be easily rediscovered by AI. Quick disclosure and immediate patching provide better protection, even with less testing. Security researchers (represented by Antoine Riard): While agreeing that embargoes for easy bugs are ineffective, Riard argued for retaining the option to fix without an advisory to avoid giving attackers extra information. Poinsot rejected this, citing the risk of panic migrations by unaffected users. Node operators and ecosystem participants: They face more frequent updates with less "bake time," which may improve security at the cost of operational burden and potential instability from hurried patches.

What to Watch

The first vulnerability classified as "easy to discover" under the new policy, and whether the disclosure process works as intended.
Whether other major open-source projects adopt similar policies in response to AI-driven vulnerability discovery.
The evolution of AI models: future models may find bugs even more easily, potentially expanding the definition of "easy" and forcing further policy changes.

Sources

Bitcoin Core changes security rules as AI makes finding code bugs dangerously easy · CoinDesk: Bitcoin, Ethereum, Crypto News and Price Data

Checked

12 claims in this story checked against 1 source: 7 supported.

Show each claim

Partly supportedAI has made bug discovery trivial.CoinDesk: Bitcoin, Ethereum, Crypto News and Price Data: “Bitcoin Core changes security rules as AI makes finding code bugs dangerously easy”
SupportedThe new rule ends months-long embargoes for vulnerabilities an AI can easily rediscover.CoinDesk: Bitcoin, Ethereum, Crypto News and Price Data: “Bitcoin Core will stop sitting on bug fixes for months when the bug is one an AI could find again”
SupportedThe rule speeds patching but gives fixes less time for testing.CoinDesk: Bitcoin, Ethereum, Crypto News and Price Data: “Maintainers said the change would speed fixes for easily rediscovered flaws, though updates may receive less testing before reaching users.”
Partly supportedThe policy replaces delayed disclosure with immediate patches and advisories for bugs that can be “found with a simple prompt to an LLM.”CoinDesk: Bitcoin, Ethereum, Crypto News and Price Data: “Those bugs get a fix copied back to every version Core still supports, shipped in a point release “as soon as possible” along with a security advisory carrying some details.”
SupportedA July audit of 390 bitcoin projects flagged nearly 5,000 potential flaws, 85 critical.CoinDesk: Bitcoin, Ethereum, Crypto News and Price Data: “A volunteer group of 16 developers calling itself the Bitcoin Red Team ran AI models across 390 bitcoin projects in about a day in late July and logged 4,962 findings , with 85 of them critical.”
SupportedBitcoin Core previously timed vulnerability disclosure by severity, with low-severity flaws disclosed two weeks after the next major version and medium- and high-severity flaws two weeks after the last affected version reached end of life.CoinDesk: Bitcoin, Ethereum, Crypto News and Price Data: “Low-severity flaws were revealed two weeks after the next major version shipped. Medium and high ones stayed under embargo until two weeks after the last affected version reached end of life.”
SupportedPoinsot wrote, “We recently experienced an increase in the number of reports for security vulnerabilities that can be easily (re-)discovered.”CoinDesk: Bitcoin, Ethereum, Crypto News and Price Data: ““We recently experienced an increase in the number of reports for security vulnerabilities that can be easily (re-)discovered,” Antoine Poinsot, the Chaincode Labs developer who wrote the change, said in his proposal.”
SupportedRiard wanted the option to ship a fix without an advisory and hold details back case by case.CoinDesk: Bitcoin, Ethereum, Crypto News and Price Data: “Antoine Riard, a longtime bitcoin security researcher, wanted Core to keep the option of shipping a fix without an advisory and holding the details back case by case.”
SupportedPoinsot wrote that “alerting users of a security vulnerability without any details could create more risk unnecessarily.”CoinDesk: Bitcoin, Ethereum, Crypto News and Price Data: ““Alerting users of a security vulnerability without any details could create more risk unnecessarily,” he wrote, pointing to “a rushed migration by users who would otherwise not be affected.””
Not in the sourcesCore Lightning told operators to disconnect because AI-generated reports overwhelmed its small team.CoinDesk: Bitcoin, Ethereum, Crypto News and Price Data: “Core Lightning, one of the main Lightning implementations, told operators to disconnect after AI-generated reports to its small team turned out to be real.”
Partly supportedThe most costly incident was a Coldcard bug that enabled thieves to drain 594 bitcoin, worth nearly $114 million, in 25 minutes across four sweeps on July 30.CoinDesk: Bitcoin, Ethereum, Crypto News and Price Data: “A bigger loss came from a bug no AI needed to find. A flaw in Coldcard hardware-wallet firmware, dormant since 2021, let attackers drain 594 BTC in 25 minutes on July 30, and losses neared $114 million across four sweeps.”
Not in the sourcesPoinsot’s proposal noted that a secret any model can regenerate in one prompt was never going to protect users for long.CoinDesk: Bitcoin, Ethereum, Crypto News and Price Data: “No bug has been flagged under the new rule yet, but a secret any model can regenerate in one prompt was never going to protect users for long.”