Author: Shaurya Malwa
Compiled by Deep潮 TechFlow
DeepChaio Summary: Bitcoin Core is no longer hiding vulnerabilities that can be easily found even without large models—AI scanning has uncovered three public flaws within a week, one of which allows strangers to crash nodes on Windows and Mac by exhausting their memory. For node operators, the pace of upgrades has been rewritten by LLMs: whereas medium-severity vulnerabilities could previously be left unpatched for a year, now patches must be rapidly applied across all versions and publicly disclosed for vulnerabilities that are “easily discoverable.”

Bitcoin Core has decided to stop withholding vulnerabilities that AI can also find, and has already disclosed three of them—one of which allows strangers to crash nodes running on Windows and Mac computers.
This vulnerability was initially flagged by an AI scan on October 23: any computer connected to a Bitcoin node running Windows, macOS, or BSD could flood the node with data until its memory was exhausted and it crashed. Developers confirmed the issue the same day, released a fix on October 27, and publicly disclosed the details on Wednesday along with two other vulnerabilities identified by the scan.
Nodes are software responsible for verifying and relaying each Bitcoin transaction, and Bitcoin Core is the program most nodes run. Until last week, Core developers would keep such medium-severity vulnerabilities secret for up to a year to give operators time to upgrade before revealing them to attackers.
Starting October 22, the policy for vulnerabilities classified as "easy to discover" has changed—because large language models have made finding such vulnerabilities inexpensive. Related fixes will now be promptly deployed to all still-supported versions, accompanied by public security advisories, as previously reported by CoinDesk.
All three vulnerabilities were identified through an automated scan using a large language model to analyze the Bitcoin Core codebase. Each finding was then verified by human contributors before being submitted to the project’s security team.
The second vulnerability explains why Core changed its rules. The scan results were reported to the project on September 15, and six days later, another developer independently discovered the same flaw: it allowed a connected node to claim it was running an outdated protocol version and then stall. The announcement stated that affected nodes could accumulate messages up to 2 MB per hour per inbound connection and up to 5 MB per hour per outbound connection.
The third vulnerability was rated as low risk and occurred three times total: once during a scan, and twice from reports by developers on August 23 and October 23. Miners could construct blocks that cause the files prepared by nodes for "reorganization"—the process by which the network replaces recent blocks with an alternative set of competing blocks—to overflow.
On a 64-bit machine, this type of attack would consume approximately one month's worth of the network's total block production. On Windows and 32-bit systems, nodes may crash outright, and the advisory states that "restarting will not help." Nodes running on 32-bit machines or storing blocks on older FAT32 drives may remain vulnerable even after patching.
Meanwhile, the release notes state that three additional medium- and high-severity vulnerabilities, already fixed in version 30.0, will be disclosed in two weeks.



