Bitcoin Core Discloses 3 AI-Found Bugs That Crash Bitcoin (BTC) Nodes
Bitcoin Core disclosed three AI-found vulnerabilities, one letting remote attackers crash BTC nodes on Windows and macOS, with three more due in two weeks.
AI SummaryAI
- Bitcoin Core disclosed three AI-found vulnerabilities on Oct. 28, one week after adopting its new disclosure policy.
- One flaw lets a remote attacker crash Bitcoin nodes running on Windows, macOS or BSD.
- The crash bug was flagged by an AI scan on Oct. 23 and fixed on Oct. 27.
- A second bug let peers stall a node, filling memory at up to 5MB per hour per outgoing connection.
Three Bugs, One New Policy
Bitcoin (BTC) Core has published the details of three vulnerabilities that an artificial intelligence scan uncovered in the node software, among them one that lets a remote machine crash a Bitcoin (BTC) node running on Windows, macOS or BSD. The write-ups went live on Wednesday, less than a week after the project adopted a new disclosure policy on Oct. 22 for bugs it judges easily discoverable now that large language models have cut the cost of finding them. The project's security advisories list all three, with the memory-exhaustion flaw described in full. That crash bug was flagged by the automated scan on Oct. 23 and confirmed by developers the same day. Fixes shipped on Oct. 27, and the advisory for the memory bug followed on Oct. 28 alongside two other findings the same scan had turned up. A node is the software that checks and relays every transaction across the Bitcoin network, and Bitcoin Core is the implementation most operators run, so a flaw in it is a network-wide matter. The Windows, macOS and BSD bug let any connected computer flood a node with data until it exhausted memory and went down. Until last week, Core developers kept medium-severity flaws of this kind confidential for as long as a year, giving operators time to upgrade before attackers learned where to look. That practice no longer applies to the easily discoverable class: fixes now ship as quickly as possible to every supported version, with a public advisory attached. All three bugs came out of one automated scan that used a large language model to comb through the codebase, and each finding was verified by a human contributor before it reached the security team. Nothing in the disclosed material touches Bitcoin price dynamics or the consensus rules; the exposure sits in resource handling on individual machines.
Update (29 Oct 14:20 UTC): Separate security research published the same week points to a longer-horizon risk to Bitcoin's cryptography itself: researchers at quantum computing firm Alice & Bob estimate that about 29,588 physical qubits, roughly 17 times fewer than Google's March estimate, could run Shor's algorithm against…
The second bug is the clearest argument for the policy change. The scan's finding reached the project on Sept. 15, and six days later a developer who had seen no report found the same flaw independently, a sequence that makes long secrecy look futile when a machine can rediscover a bug inside a week. The flaw let a connected peer claim to run an outdated version of the protocol and then stall, with messages piling up in the victim's memory at up to 2MB an hour for each incoming connection and 5MB an hour for each outgoing one, the advisory states. The third flaw, rated low severity, surfaced three separate times: in the scan, and in two independent reports from developers, in August and again on Oct. 23. A miner could craft blocks that overflow the files a node keeps for a reorganization, the process by which the network swaps a few recent blocks for a competing set under the proof of work rules that order the chain. On a 64-bit machine the attack would need roughly a month of the network's entire block production. On Windows and 32-bit systems the node can fail outright, and restarting does not clear the failure, the advisory says. Nodes on 32-bit hardware, or storing blocks on older FAT32 drives, remain exposed even after the fix. The release notes add that three more medium- and high-severity bugs, already fixed in
Bitcoin (BTC) Core version 30.0, will be disclosed in two weeks. Operators on supported 64-bit builds clear the recorded exposures by upgrading to the patched releases; the constraints that leave any residue are hardware and storage, not code.
New details from the advisories fill in the technical picture behind the crash bug: Bitcoin Core normally caps how much queued data each peer can hold and pauses reading when it overflows, but the Windows, macOS and BSD builds use a different connection-monitoring mechanism that keeps reading past the pause when it receives TCP urgent data, letting memory grow without limit. Linux nodes are not affected. The patched releases are versioned 29.5, 30.4, 31.2 and 32.1, with 31.2 and 32.1 out on Oct. 27 and the two older backports on Oct. 28. The scan behind all three findings was Project Loupe, an automated effort led by Spiral's Elias Rohrer, with Gregory Sanders verifying the findings before reporting them. Antoine Poinsot, who proposed the Oct. 22 policy change, wrote that LLMs have sharply lowered the bar for rediscovering certain bug classes, making long embargo periods ineffective while delaying patched releases.
The 32.1 release notes add a detail the advisories omit: version 32.0, although tagged, was never fully announced or released, and the notes credit "Anthropic Security" among contributors without specifying what it worked on. The advisories themselves do not name the AI model behind the scan. Project Loupe, the scanner behind all three findings, was launched in May by Spiral and Block as a free vulnerability scanner for open-source
Bitcoin (BTC) projects, with Bitcoin Core among its first test projects.
The developers rated the memory-exhaustion flaw as medium severity, a judgment the advisories carry and that fits the machine-assisted discovery path: the team also classed the bug as easily findable, the category the revised policy covers. A Japanese-language report on the disclosure spells out what a crash means operationally: an attacked node stops confirming and relaying transactions, while the network as a whole carries on, since the many remaining nodes keep verifying blocks and payments independently. That same account places the official write-up on Bitcoin Core's website on Oct. 28, five days after the Oct. 23 scan flagged the issue, with patched builds already available. For affected operators on Windows, macOS or BSD, the stated remedy is simply updating to the fixed version.
Separate security research published the same week points to a longer-horizon risk to Bitcoin's cryptography itself: researchers at quantum computing firm Alice & Bob estimate that about 29,588 physical qubits, roughly 17 times fewer than Google's March estimate, could run Shor's algorithm against wallet signatures in roughly 62 minutes. That is too slow to reliably strike a transaction before its 10-minute confirmation window closes, but more than six million bitcoins sit at addresses whose public keys are already exposed, giving a future attacker more time to work back to a private key. Senior cryptanalyst Dahmun Goudarzi called the findings theoretical, with hardware improvements and falling resource requirements expected to converge in the early 2030s, and said migration to quantum-resistant security should start now because replacing signatures across the network could take years.
No Attacker Use On Record
COINOTAG's reading is that the policy shift re-prices secrecy: once an LLM can reach a bug, the project treats it as public regardless of who might have kept it private, and the advisory record now backs that stance with dates rather than intentions. The immediate risk concentrates on operators running 32-bit systems or FAT32 storage, since the patch does not cover them. Mining operations were never in scope; the flaws sit in node software, not in mining hardware or rewards. Nothing in the sources records a node crashed by an attacker in the wild, and no incident has been reported since the fixes shipped. That first confirmed report of a crashed node is the observation that would turn this patch notice into an incident; it has not been made (as of 08:20 UTC).
AI-generated, AI-reviewed, under COINOTAG editorial oversight.

