Bitcoin Core Discloses Three AI-Found Bugs Under New Fast-Track Rules

The first real test of Bitcoin Core's new vulnerability tier arrived fast, and the system held.

Key takeaways

• Bitcoin Core published three vulnerability advisories on October 28, the first ever under its new "easily discoverable" disclosure tier, including a medium-severity crash bug allowing any connected peer to exhaust memory and kill a Windows, macOS, or BSD node; Linux nodes are unaffected.
• Two of the three bugs were surfaced by Project Loupe, an automated LLM scanner built by Elias Rohrer at Spiral, with Gregory Sanders validating the findings and reporting them to the Bitcoin Core security team; discovery to patched release took five days.
• Fixes are live in versions 29.5, 30.4, 31.2, and 32.1, node operators on Windows, macOS, or BSD should update immediately; no exploitation has been observed.

Bitcoin Core disclosed three vulnerabilities on October 28, 2026, the first ever published under its new "easily discoverable" disclosure tier, which fast-tracks fixes and simultaneous public advisories for bugs that automated tools can find quickly. The lead flaw is medium-severity: a malicious connected peer can send TCP urgent (out-of-band) data, triggering unbounded memory growth until the node runs out of memory and crashes. Windows, macOS, and BSD nodes are affected. Linux is not, because Linux uses poll() rather than select().

The second flaw is also rated medium-severity. A peer pretending to use a pre-BIP31 protocol (which lacks modern ping messages) can avoid disconnect timeouts and let queued announcements grow without bound, producing a separate memory-exhaustion path. A third, low-severity flaw involves undo-file overflow that could corrupt blockchain rollback data under specific reorg-based attack conditions; the advisory rates exploitation risk as small. Patched versions, 29.5, 30.4, 31.2, and 32.1, are available now at bitcoincore.org.

Five Days From Discovery to Fix

The timeline on the lead vulnerability is tight. On October 23, Project Loupe, an automated LLM-powered code scanning tool built by Elias Rohrer at Spiral, identified the TCP crash bug. Gregory Sanders validated the finding and notified the Bitcoin Core security team the same day. On October 26, Chaincode Labs developer Antoine Poinsot opened and merged PR #36665 to fix this issue. Versions 31.2 and 32.1 shipped on October 27. Versions 29.5 and 30.4 followed on October 28, alongside public disclosure. No CVEs appear to have been assigned as of disclosure.

That five-day window from identification to patched release is the new policy working exactly as Poinsot argued it should. The new disclosure tier, merged on October 22, was built on a specific premise: bugs that LLMs can find quickly should not sit under a year-long embargo, because the embargo only delays the fix while doing nothing to stop a motivated attacker from running the same scan. Poinsot made this case in the GitHub PR proposing the change, arguing that long embargoes delay the availability of fixed releases without meaningfully reducing attacker opportunity.

What This Actually Means for Node Operators

A crashed node is not a stolen wallet. None of these bugs allow an attacker to move funds. But a node that has been forced offline has stopped verifying and relaying transactions, which matters if that node is the trust anchor for your payments or your Lightning channel state. Self-sovereignty is only as strong as the software version running it.

The LLM-as-auditor dynamic is now permanent, and that has second-order consequences worth tracking. Project Loupe found real, exploitable bugs in the most security-reviewed open-source codebase in existence. The cost of attack-surface discovery has dropped. Bitcoin Core's developers read that correctly and updated policy before a bad actor could exploit the gap. That's a maturity signal, not a panic signal.

The no-CVE approach is a data point to watch. CVEs are how enterprise IT environments prioritize patching across infrastructure. Bitcoin node operators running in institutional settings may not receive the alert through standard channels without a CVE attached. The Bitnodes version distribution in the 30 days following this disclosure will show whether that gap costs operator response time in practice. That's the trackable question from here.

The broader context: in 2026, TFTC covered active attacks on unpatched Core Lightning nodes after AI-generated vulnerability reports proved genuine. The pattern is consistent. AI scanning finds real bugs. The question is whether the disclosure and patching infrastructure responds faster than adversarial actors can weaponize the same tools. This week's disclosure suggests it can.

What to Watch

The falsifiable test of the new policy is in the Bitnodes data. If a significant share of the node population remains on unpatched versions 30 days out, the fast-disclosure timeline may have outrun operators' upgrade cadence, trading the risk of LLM rediscovery for the risk of mass exposure mid-window. If version uptake tracks normally, the policy upgrade holds. Watch that distribution. And update your node.

Sources

• Bitcoin Core advisory: TCP urgent data OOM
• Bitcoin Core release announcements
• GitHub PR #1296, darosior publishes three advisories