Bitcoin Core disclosed three security vulnerabilities on October 28, 2026, the first bugs published under a new "easily discoverable" fast-track disclosure tier adopted six days earlier. Two of the three flaws were identified by Project Loupe, an automated LLM-powered scanner built by Spiral, Blo...
"The embargo period achieves little but it delays the availability of a fixed release. We are therefore updating our policy disclosure to accommodate those." — Antoine Poinsot, Bitcoin Core Developer
Bitcoin Core disclosed three security vulnerabilities on October 28, 2026, the first bugs published under a new "easily discoverable" fast-track disclosure tier adopted six days earlier. Two of the three flaws were identified by Project Loupe, an automated LLM-powered scanner built by Spiral, Block's open-source Bitcoin division. Discovery-to-patch took five days.
The policy change, merged October 22, eliminates long embargoes for vulnerabilities that an attacker can rediscover with a single AI prompt. It marks the first formal acknowledgment by a major open-source infrastructure project that LLM-era tooling has collapsed the asymmetry between security researchers and attackers, rendering secrecy-based disclosure timelines counterproductive for an identifiable class of bugs.
The broader context: a volunteer Bitcoin Red Team audited 390 Bitcoin-related projects in 27.5 hours on August 4-5, filing 4,962 findings — 85 critical, 635 high-severity — largely through AI-assisted scanning. That sprint was triggered by the Coldcard hardware wallet RNG flaw, which enabled the theft of 1,816 BTC ($116 million) starting July 30.
All three bugs affect Bitcoin Core's networking stack. None enable fund theft. All enable denial-of-service against node operators.
Vulnerability 1: TCP Urgent Data Memory Exhaustion (Medium)
A malicious peer sends TCP out-of-band (urgent) data to a Bitcoin Core node. On systems using the select() socket-monitoring backend — Windows, macOS, and BSD — this bypasses the per-peer data queuing limit. The node's memory grows without bound until it crashes. Linux nodes are unaffected because they use poll() instead of select().
Affected versions: Bitcoin Core 29.x through 32.x (pre-patch). Fixed in: 29.5, 30.4, 31.2, 32.1.
Vulnerability 2: Paused Peer Memory Growth (Medium) A peer mimics the pre-BIP31 protocol, which lacks modern ping messages. This allows it to evade disconnect timeouts while queued transaction announcements accumulate unbounded in memory. The result is the same: memory exhaustion and node crash.
Vulnerability 3: Undo-File Overflow (Low) Under specific blockchain reorganization conditions, an attacker can trigger an overflow in the undo-file data structure, corrupting rollback data. The exploitation window is narrow and the Bitcoin Core team assessed the practical risk as small.
Timeline:
No CVE numbers were assigned as of disclosure.
Project Loupe is a free vulnerability scanning service built by Elias Rohrer, Jordan Mecom, and Thomas Kilbride at Spiral, published May 12, 2026. It runs LLM-based analysis across open-source Bitcoin codebases and reports only vulnerabilities backed by demonstrable test cases — eliminating the false-positive noise that plagues most automated security tools.
Initial test partners include Bitcoin Core, BDK, LDK, rust-bitcoin, Cashu, Blockstream Jade, bitcoinj, and SRI. Spiral does not disclose which specific LLM models Loupe uses, stating only that the tool is designed to stay current as the model landscape evolves.
Two of the three October 28 disclosures were surfaced by Loupe. The third — the undo-file overflow — was identified through a separate review process.
The significance is structural. Bitcoin Core's C++ codebase is among the most scrutinized open-source projects in existence, with 18 years of peer review, multiple prior professional audits, and a security-conscious development culture. That an LLM scanner found exploitable bugs in five days of operation against this codebase suggests that the technique scales to less-reviewed projects with substantially larger attack surfaces.
Spiral frames the tool as an equalizer. According to its announcement: "The asymmetry between attackers and maintainers shouldn't decide bitcoin's security." Most critical Bitcoin infrastructure is maintained by small teams or individual developers without enterprise-grade security budgets. Loupe provides continuous scanning at no cost, funded by Block.
On October 22, Bitcoin Core merged a revision to its vulnerability disclosure policy, introduced by developer Antoine Poinsot via the Bitcoin Development Mailing List. The change creates a new tier — "easily discoverable" — defined as flaws "that can be identified with little effort," with LLM prompts cited as the exemplary case.
How the tiers now work:
| Tier | Disclosure Timing | |------|------------------| | Low severity | 2 weeks after next major release | | Medium/High severity | 2 weeks after affected release reaches end-of-life (~1 year) | | Easily discoverable | Immediate fix across all maintained versions + simultaneous advisory | | Critical | Ad hoc, case-by-case |
The rationale is direct: if an attacker can find the bug with a single prompt, a 12-month embargo protects nothing. It delays the fix without preventing discovery by motivated adversaries.
The October 28 disclosures were the first real-world application of this tier. The policy is not limited to AI-discovered issues; it covers any vulnerability where discoverability is trivially low, regardless of the tool used.
A notable trade-off: accelerated patches receive less testing before reaching users. Node operators may face a choice between running a less-tested patch immediately or remaining exposed to a publicly disclosed vulnerability. Bitcoin Core has not published guidance on how operators should weigh this trade-off.
The policy change did not occur in a vacuum. On August 4-5, 2026, a volunteer group called the Bitcoin Red Team conducted a 27.5-hour sprint across 390 Bitcoin-related software projects, funded by a nearly $40,000 OpenSats grant. The team comprised 16 human researchers and 3 automated agents — 17 contributors total.
Results:
Before the August sprint, the Red Team had scanned roughly 150 repositories through smaller-scale efforts, producing over a dozen private disclosures through responsible disclosure processes. The team has announced plans to open-source its scanning harness.
The data point that matters: 91% automated intake. The human researchers primarily triaged and validated AI-generated reports rather than identifying bugs manually. This inverts the traditional audit model, where human experts do primary discovery and tools assist. Here, the tools do primary discovery and humans validate.
The Red Team sprint was triggered by a specific incident. On July 30, 2026, attackers began draining wallets generated by Coldcard hardware wallets running firmware version 4.0.1, shipped in March 2021. A build configuration error set the macro MICROPY_HW_ENABLE_RNG to zero, causing the device's random number generator to fall back to MicroPython's deterministic Yasmarang algorithm instead of the STM32 hardware RNG.
The fallback relied on the device's microcontroller identifier and system timing values — neither cryptographically secure. The result: wallet seeds were predictable.
According to TRM Labs, attackers brute-forced the weak seeds in 41 minutes, draining 1,816 BTC from more than 5,200 addresses across four attack waves. At the time of the attack, that represented approximately $116 million. By mid-September, blockchain analytics firm TRM Labs classified it as the largest hardware wallet exploit on record.
The flaw had existed undetected for over five years. A single build flag, set incorrectly in 2021, undermined the cryptographic foundation of every wallet generated on affected devices. The incident demonstrated that critical vulnerabilities in Bitcoin infrastructure can persist for years under manual review processes — and that the ecosystem's security posture required a structural response.
The three disclosed vulnerabilities do not enable fund theft. They enable node crashes. The economic impact is indirect but real.
A self-sovereign user who runs their own Bitcoin Core node for payment verification loses that verification capability if the node is forced offline. Lightning Network operators face a more acute risk: if a node goes down, channel counterparties can attempt to broadcast outdated channel states. The time between node crash and restart becomes a window of financial exposure.
For mining pool operators, a crashed node means interrupted block template construction and potentially missed blocks. At current block rewards of 3.125 BTC (approximately $250,000 per block), even brief downtime carries material cost.
The simultaneous disclosure model creates a timing problem. When a fix and its corresponding vulnerability advisory publish at the same time, every unpatched node becomes a known target immediately. Node operators who do not update within hours of disclosure face adversaries who now have both the vulnerability details and confirmed-affected version numbers.
According to Bitnodes data, approximately 20,000 reachable Bitcoin Core nodes operate globally. The share running on Windows, macOS, or BSD — the platforms affected by the TCP vulnerability — is estimated at 15-25%, implying 3,000-5,000 nodes were exposed prior to patching.
The October 28 disclosures represent a structural shift in how Bitcoin's core infrastructure handles security. The traditional model — long embargoes, manual audits, private disclosure timelines measured in months — assumed that obscurity provided meaningful protection. LLM-powered scanning tools have invalidated that assumption for an identifiable class of vulnerabilities.
The data from the Bitcoin Red Team sprint quantifies the scale: 4,962 findings across 390 projects in 27.5 hours, with 91% generated by automated tools. The Coldcard incident quantifies the cost of the old model: $116 million lost to a five-year-old build flag error that manual review never caught.
Bitcoin Core's response — creating a fast-track disclosure tier, accepting AI-found bugs as first-class reports, and publishing patches with same-day advisories — is pragmatic. It acknowledges that the security landscape has changed and adjusts process accordingly.
The open question is whether the broader Bitcoin ecosystem follows. Coldcard's firmware, Lightning implementations, hardware wallet stacks, and the hundreds of smaller projects identified by the Red Team audit all face the same asymmetry: attackers can now run the same LLM scans that security researchers use. The projects that adopt continuous AI-assisted scanning will find their bugs first. The projects that do not will have their bugs found for them.