NewsCryptoBitcoin Core Patches Node-Stall Bug First Surfaced by an AI Scan

Bitcoin Core Patches Node-Stall Bug First Surfaced by an AI Scan

Author: Coindoo·

Key Takeaways

  • •CVE-2026-117348 allowed attackers to stall Bitcoin Core nodes for over 12 seconds at a time by sending orphan transactions with thousands of missing parents and inv messages repeating the same wtxid up to 50,000 times, though the flaw affected node performance rather than validation integrity or wallet security.
  • •The bug was first surfaced on July 8 by an automated LLM scan at Project Loupe and independently rediscovered on August 4 by a second researcher, placing it in Bitcoin Core's 'easily discoverable' category and triggering public disclosure alongside the fixed releases.
  • •Fixes are available in Bitcoin Core 30.5 and 31.3, both released November 20, and in version 32.1, which launched on October 27 with the patch already included.
  • •Nodes running versions 29.0 through 29.5 on the end-of-life 29.x branch have no patch available and must upgrade to 30.5, 31.3, or 32.1 to close the vulnerability.
  • •Services that depend on a single Bitcoin Core instance, such as merchant payment servers, wallet backends, or Lightning nodes, could fall behind during a stall, so a payment may confirm on-chain before the service registers it.
Bitcoin Core Patches Node-Stall Bug First Surfaced by an AI Scan

Bitcoin Core developers published a security advisory for CVE-2026-117348 on November 20, disclosing a medium-severity flaw that could stall affected nodes for more than 12 seconds at a time. The bug was first surfaced by an automated large language model scan and was independently rediscovered by a second researcher weeks later — a pattern that placed it in the project's "easily discoverable" category and moved it onto a fast-track public disclosure.

The fixes ship in Bitcoin Core 30.5 and 31.3, both released on November 20, and in version 32.1, which launched on October 27 with the patch already included. Nodes running the end-of-life 29.x branch have no patch available and must upgrade to close the vulnerability.

How an AI Scan and a Second Researcher Found It

The first report arrived on July 8 from Project Loupe, Elias Rohrer's research effort at Spiral. The flaw surfaced in an automated LLM scan, and Bitcoin Core contributor Gregory Sanders validated the finding before taking it to the project's security team. On August 4, a researcher known as Stutxo reported the same issue independently.

Two unrelated parties finding one bug within four weeks is the risk the "easily discoverable" category is designed to address. Under Bitcoin Core's disclosure policy, such bugs receive a public advisory alongside the fixed releases, without the usual severity-based waiting period. By comparison, a high-severity mining-template bug disclosed earlier this month stayed under embargo for 396 days after its fix. Coindoo has covered why the project moved to the faster track, as well as the earlier mining-template disclosure.

What the Bug Did Inside a Node

An orphan transaction spends an output from an earlier transaction that the node has not yet received. That alone does not make it invalid. The node keeps the transaction in a holding area, known as the orphanage, until the missing parent transactions turn up.

Starting with version 29.0, when a peer announced an orphan by its wtxid, the node listed the orphan's parents and checked each one against its mempool and its filters of recently seen transactions. Peers announce transactions in inv messages, which can carry up to 50,000 entries, and the software did not skip duplicates.

That gave an attacker a straightforward way to waste a node's time. According to the advisory, the attacker could send an orphan with thousands of missing parents, then send inv messages repeating its wtxid 50,000 times. Every repetition triggered the full parent check again. Each such message could stall the node for more than 12 seconds, and during that window the node processed nothing from other peers, blocks included.

The advisory describes the stall as temporary and does not mention any exploitation. Nothing in the advisory suggests the flaw let invalid transactions through validation or exposed wallet keys. It affected how fast a node could keep up, not what the node accepted.

The Fix

Sanders opened the fix as pull request #35878 on August 4, and it was merged two days later. A patched node handles each announced transaction once per inv message, however many times the message repeats it. The 31.3 release notes list the change as "process unique tx INVs only."

Who Got the Fix, and When

How the patch reached users depends on which branch they run:

  • 32.x — patched from launch. Version 32.1, released October 27, shipped with the fix. Version 32.0 was tagged but never released.
  • 31.x — fixed November 20. Versions 31.0 through 31.2 are affected. Operators should upgrade to 31.3.
  • 30.x — fixed November 20. Versions 30.0 through 30.4 are affected. Operators should upgrade to 30.5.
  • 29.x — no fix planned. Versions 29.0 through 29.5 are affected. The branch reached end of life on October 27.

Anyone on 32.x has been running patched code since the branch's first public release, weeks before the advisory went out. The 30.x and 31.x fixes were backported on November 17 and released alongside the advisory, so users of those branches learned about the flaw and received the patch at the same moment.

Under Bitcoin Core's maintenance policy, the project covers the three latest major versions, and branches past end of life generally do not receive security fixes. That places 30.x, 31.x and 32.x inside the supported window today, with 30.x the oldest branch still covered and the next in line to age out as new major versions arrive. The project recommends running the latest maintenance release of the most recent major version an operator can upgrade to.

Which Setups Feel a Stall First

While a node is stalled, it is not reading new blocks or transactions from its peers. The rest of the network carries on. Only that node's view falls behind, and it catches up once processing resumes.

The exposure depends on what relies on the stalled node. A merchant payment server, a wallet backend or a Lightning node that takes chain data from a single Bitcoin Core instance sees whatever delay that instance sees, so a payment can confirm on-chain before the service registers it. Operators who run several nodes or keep a fallback data source are better placed to ride out a stall on one of them.

Checking That a Node Is Patched and Keeping Up

A node that is running is not necessarily patched or in sync. After upgrading through the official release instructions, three checks confirm the fix is live and the node is current:

  1. Confirm the running version. bitcoin-cli getnetworkinfo returns a subversion field such as /Satoshi:31.3.0/. Query the live process, because a node that has not restarted is still running the old binary.
  2. Confirm the node has caught up. In bitcoin-cli getblockchaininfo, blocks should match headers once the node is in sync.
  3. Confirm dependent services see the same chain. Compare the node's tip height against an independent source and against any payment, wallet or Lightning software that reads from it.

Operators still on 29.x will not find a patched build to download. Closing this gap means moving to 30.5, 31.3 or 32.1.

Source: Coindoo