Bitcoin Core maintainers disclosed three security vulnerabilities discovered by artificial intelligence, including one flaw that lets a malicious peer remotely crash nodes running on Windows and macOS, with fixes shipped across versions 29.5, 30.4, 31.2 and 32.1.
By Sarah Park | October 29, 2026
If you run a Bitcoin node, update it. If you simply own bitcoin, the story still matters: it is the clearest evidence yet that AI tools are now auditing the software securing hundreds of billions of dollars in value, and that the maintainers of the network’s reference software are racing to fix what those tools find before anyone else does.
The Hook: Three Flaws, Found by a Language Model
On October 28, 2026, Bitcoin Core published security advisories describing two medium-severity memory-exhaustion vulnerabilities and one low-severity file-overflow flaw. All three were initially identified through Project Loupe, an automated scanning initiative that uses large language models, developed by Elias Rohrer at Spiral. Gregory Sanders then validated the findings and alerted Bitcoin Core’s security team.
The most immediately exploitable flaw affects nodes running on Windows, macOS, and BSD. A malicious peer can send specially crafted network data that bypasses safeguards limiting incoming data, causing memory consumption to climb until the node crashes. Linux systems are unaffected by this particular flaw because they use a different mechanism for monitoring network connections.
On-Chain Evidence: How the Attacks Work
The second medium-severity flaw sits in how Bitcoin Core communicates with older peers that do not support a network feature introduced under Bitcoin Improvement Proposal 31. A malicious peer can exploit the mechanism to accumulate outgoing messages indefinitely, growing memory use by up to 2 megabytes per hour per inbound connection and 5 megabytes per hour per outbound connection until a node runs out of memory and drops offline.
The third vulnerability involves the files Bitcoin Core uses to reverse transactions during blockchain reorganizations. Under highly unusual circumstances, a miner could generate enough transaction data to overflow file-position limits, potentially corrupting reorganization records and forcing affected nodes to shut down. Bitcoin Core rated it low severity because exploiting it on conventional 64-bit systems would require mining thousands of specially constructed blocks and engineering a precise reorganization.
- July 8: the earliest of the three flaws was discovered by the AI scanner.
- September 15: the second vulnerability was found.
- October 23: the Windows and macOS memory-exhaustion flaw was identified.
- October 26: patches were completed, with releases shipped over the following two days before public disclosure.
The advisories did not report any active exploitation in the wild. None of the flaws touch private keys or allow coin theft; the risk is disruption, draining the nodes that validate transactions and relay blocks across the network.
The Core Conflict: Speed of Disclosure Versus Speed of Attackers
The accelerated response was deliberate. Bitcoin Core’s security policy now calls for immediate disclosure once fixes become available for easily discoverable weaknesses, including those detectable through relatively simple AI-assisted scanning. The logic is blunt: if a language model can find the bug, so can anyone else’s language model. Sitting on patches while users run vulnerable software is no longer the conservative choice it once seemed.
But the patches do not erase every exposure. Bitcoin Core warned that 32-bit systems and nodes storing blockchain data on FAT32 file systems remain susceptible to the file-overflow issue in specific conditions. The 32.1 release notes also confirm that version 29.x and older branches have reached end of life, meaning operators on those releases get no future security updates even if they install the 29.5 patch. Moving to maintained versions requires macOS 14 or newer and Windows 10 version 1903 or newer for Bitcoin Core 32.1, and migrations from obsolete releases may need extra time for data-directory moves.
Market Implications for Everyday Investors
For holders, the reassuring part is what did not happen: no funds were at risk, and the network kept running. The unsettling part is the new baseline. The same AI wave that is reshaping every industry is now a permanent part of offense and defense in crypto security. Researchers have already used AI to find a flaw in the XRP Ledger that could theoretically mint trillions of tokens, and Bitcoin is getting the same treatment. Expect faster disclosure cycles, more advisories, and more pressure on anyone running stale software.
The Verdict
This episode is a net positive for Bitcoin’s long-term health: weaknesses were found by friendly scanners, fixed quickly, and disclosed honestly. The burden now shifts to operators, exchanges, and hobbyists running nodes to actually install the updates. If you rely on a third party to hold your bitcoin, the practical question is simple: ask whether their infrastructure runs maintained software. In an era where AI finds bugs in days, being three versions behind is not caution, it is risk.
The cryptocurrency market remains highly volatile. This article is for informational purposes only and does not constitute financial advice.
updated my node within an hour of the advisory. the remote crash on windows is nastier than low severity sounds, a single hostile peer can trigger it
which build are you on? running 29.x here and the notes were light on whether anything before 29.5 is affected at all
Bjorn, everything before 29.5 is affected, the advisory lists fixes in 29.5, 30.4, 31.2 and 32.1. any older branch and you are running the memory exhaustion bugs
an AI found flaws humans missed for years in software securing a trillion dollars. the audit wars just went automated, expect a flood of these advisories
fixes landed in 29.5, 30.4, 31.2 and 32.1. so anything older than 29.5 should be treated as affected, the advisory was light on that distinction
file overflow bugs on windows nodes are how you set up bigger problems later. update now, complain later
project loupe finding memory exhaustion bugs before humans did is a bigger story than the bugs themselves. llms auditing the software securing billions
the rohrer team still validated every finding before disclosure though. llm flags candidates, humans confirm. that is a sane pipeline, not automated audits shipping straight to advisories
double edged though. if an llm can find these flaws cheaply, so can people with worse intentions. the disclosure race speeds up for everyone
this is the part everyone keeps skipping. project loupe is public now, so every security team and every attacker knows llm fuzzing works on bitcoin core. disclosure race speeds up on both sides
the llm that found these was apparently prompted by the maintainers themselves. imagine the same sweep run by someone who quietly reports to the wrong bug bounty
^ this is the uncomfortable part. the same sweep costs a few dollars to run and most node ops update slower than researchers disclose
updated to 29.5 the minute i saw the advisory. if you’re on windows especially, a remote crash from a malicious peer is not theoretical
^ exactly. patch latency is the real attack window