Luke Dashjr Disputes Claim That Only AI Could Find Coldcard Flaws

Luke Dashjr Rebuts Coinkite’s Artificial Intelligence Defense

Veteran Bitcoin Core developer Luke Dashjr published an extensive technical breakdown on X examining the firmware flaws found in Coinkite’s Coldcard hardware wallets. Dashjr’s analysis questioned Coinkite’s assertion that external attackers required cutting-edge artificial intelligence to detect foundational weaknesses in its random-number generation architecture. Addressing the narrative, Luke Dashjr stated, “But the claim that only a ‘genius AI’ could find issues with the RNG is a blatant lie,” emphasizing that standard review practices should have revealed multiple structural irregularities.

In his technical assessment, Dashjr identified an unresolved TODO note within rng_init, the specific routine responsible for initializing the Mk4 Coldcard’s RNG. Dashjr noted that certain STM32 microcontrollers require developers to discard the initial sample upon initialization. Furthermore, Dashjr observed that the driver failed to account for the chip’s internal seed-error and clock-error status bits, despite STM32 reference manuals explicitly prescribing extra post-processing steps for cryptographic operations.

Code vulnerabilities extended further into runtime validation. While the random_buffer function was designed to catch duplicate RNG output, direct calls targeting rng_get lacked equivalent safeguards. Dashjr stated that rng_get_or_fault cannot be treated as secure until developers definitively prove the random_buffer check is redundant. He also questioned why the secure element’s entropy failed to maintain a full 256 bits if the STM32 hardware output became deterministic. Classifying the remediation as essentially a novel, unproven RNG requiring extensive testing, Dashjr cautioned that the patch only deactivated the obsolete pseudo-random number generator (PRNG) at compile time rather than applying runtime assertions. Dashjr warned that improper RNG timeouts could trigger undefined behavior or firmware crashes, presenting a risk of bricking slower devices.

Anatomy of the $140 Million Coldcard Seed Generation Breach

The core vulnerability originated from an engineering shift in March 2021, when Coinkite relocated seed generation away from the dedicated hardware random-number chip to Yasmarang, an internal software tool driven merely by device serial numbers and processor timers. Consequently, Mk2 and Mk3 Coldcard units produced seeds featuring roughly 40 bits of genuine entropy—substantially lower than the 128-bit baseline established by the BIP-39 standard. This catastrophic drop in randomness empowered attackers to systematically compute and recreate vulnerable wallet private keys offline.

On July 30, 2026, an automated exploit struck vulnerable Coldcards, draining roughly 594 BTC (worth approximately $38 million) across roughly 500 individual wallets within a single 25-minute window. While whitehat operations managed to intercept some balances, total cumulative losses stemming from the breach are estimated to exceed $140 million. In an official technical briefing issued the day the attack occurred, Coinkite claimed the firm was “unaware of the bug until today.” The company speculated that an unauthorized party may have deployed AI tools against its open-source firmware repo to isolate the flaw, after an internal AI-assisted review conducted weeks prior missed the defect.

To mitigate the crisis, Coinkite rolled out firmware version 5.6.1 for the Coldcard Mk4 and Mk5, as well as version 1.5.1Q for the Coldcard Q on August 20. The company stated the security updates incorporated “AI-assisted review by Kimi and other frontier models” alongside community contributions. The revised firmware introduced checks addressing several of Dashjr’s points: it monitors the hardware RNG seed-error state, automatically purges the following 12 hardware words when recoverable errors occur, institutes a boot check confirming rng_get() accesses the true hardware read path, and migrates the backup PRNG system to SHA-256 Hash_DRBG.

Why This Matters

The Coldcard compromise marks one of the most consequential hardware security failures in Bitcoin history, directly challenging user trust in cold-storage seed integrity and vendor-led security audits. The fallout has expanded beyond code reviews into legal liability. Amir Reza Naderi, a Coldcard user, has initiated a proposed class-action lawsuit in the Ontario Superior Court of Justice against Coinkite, Novak, and co-founder Peter D. Gray.

The lawsuit claims Coinkite management discovered the defective seed-generation implementation as early as May 2025 yet withheld warnings from device owners. While Coinkite has vigorously denied allegations of prior knowledge, the combination of developer disputes, multi-million dollar asset losses, and ongoing legal actions ensures prolonged regulatory and community scrutiny regarding how hardware vendors implement and audit cryptographic primitives.

Frequently Asked Questions

What caused the Coldcard hardware wallet seed-generation vulnerability?

The vulnerability stemmed from a March 2021 firmware modification that replaced the dedicated hardware random-number chip’s entropy generation with a software tool named Yasmarang. Yasmarang derived its entropy solely from hardware serial numbers and timers, reducing randomness on Mk2 and Mk3 models to roughly 40 bits instead of the standard 128 bits, which allowed malicious actors to brute-force private keys offline.

What were Luke Dashjr’s primary technical criticisms of the Coldcard RNG codebase?

Luke Dashjr highlighted an unaddressed TODO note regarding STM32 initialization samples, unmonitored seed-error and clock-error status bits, missing redundancy validations between direct rng_get calls and random_buffer, and reliance on compile-time rather than runtime checks. Dashjr also argued that standard developer auditing—not complex artificial intelligence—was more than sufficient to detect these design oversights.

What claims are being made in the Ontario class-action lawsuit against Coinkite?

The proposed class action, filed by customer Amir Reza Naderi in the Ontario Superior Court of Justice, names Coinkite, Novak, and co-founder Peter D. Gray as defendants. The complaint alleges that Coinkite was aware of the seed-generation defect starting in May 2025 and failed to warn device owners before widespread thefts occurred. Coinkite has formally denied these allegations.