Coldcard Seed Flaw Puts Security Reviews Under Scrutiny
TRM Labs linked a Coldcard seed-generation flaw to a preliminary estimate of about 1, 816 BTC stolen, worth roughly $116 million at the time of reporting. Coinkite CEO NVK has defended the company’s review history, but available accounts do not show whether earlier reviewers examined the faulty build configuration or whether anyone reported the same exploitable defect before the theft.
- TRM Labs traced the flaw to a March 2021 build configuration error.
- The estimated theft spans more than 5, 200 addresses, not necessarily that many people.
- NVK says at least 15 external efforts examined Coldcard-related systems.
- A firmware update cannot fix a seed already generated under vulnerable conditions.
How a seed-generation flaw can expose funds
A wallet seed is the secret used to derive its private keys. It must be unpredictable. If a faulty process draws a seed from far fewer possibilities than intended, an attacker may be able to search that smaller set and derive the keys.
TRM Labs traced the Coldcard issue to a build configuration error in firmware 4.0.1, released in March 2021. According to TRM, the configuration led some devices to use a weak software random-number generator instead of the intended hardware-based source of entropy, the unpredictability used to create a seed. TRM said attackers could reconstruct keys from affected seeds without physical access to the devices.
That does not mean attackers remotely took control of the hardware wallets. The reported weakness involved seed generation. If a seed was predictable enough, its keys could potentially be derived without touching the device.
TRM’s preliminary estimate, based on Galaxy Research’s running tally, put the theft at roughly 1, 816 BTC from more than 5, 200 addresses across four waves beginning July 30, 2026. The roughly $116 million figure reflects the bitcoin’s value at the time, not a fixed valuation. An address is not the same as a person or a wallet, so the address count should not be taken as a confirmed victim count. TRM described its tally as preliminary.
What the review count does and does not show
On November 9, 2026, NVK said at least 15 external reviews or research efforts had examined Coldcard-related hardware, firmware, tools, or release processes. He cited paid reviews of the Mk4 in 2022 and the Q in 2024. Available accounts do not give the full scope or findings of those reviews, or show whether they examined the faulty build configuration.
NVK also said a review associated with Rob Ham’s OpenAI Cybersecurity Harness produced 85 candidate findings before the incident, several of which led to fixes. He said the review did not identify the seed-generation defect. “Candidate findings” does not mean 85 confirmed vulnerabilities. The accounts available here do not include the review’s full report or methodology.
“The uncomfortable truth is that it took a new AI model to find what humans couldn’t.”
That is NVK’s characterization, not an established account of who first found the flaw. Available accounts do not say which model identified it, when it did so, or how the finding was verified. They also do not establish that no human had identified the same issue earlier.
Bitcoin developer Luke Dashjr disputed the implication that human reviewers had not found relevant problems. He said he had privately shared findings with Coinkite and was now disclosing them publicly. He described his own examination as a quick look, not a full audit, and cautioned that he was not claiming every issue he raised was definitely a bug or exploitable.
Other participants disagreed about what counts as meaningful scrutiny. Alex Waltz accused Coinkite and NVK of discouraging code examination. Kenneth said ongoing quality assurance differs from asking a team to review an entire codebase. Shinobi defended NVK, saying reviewers had examined the entropy system, and called for a reproducible demonstration of the suspected failure.
These efforts are not interchangeable: checking hardware randomness, reviewing release controls, briefly examining source code, and auditing the exact firmware build are different kinds of work. Without their scope and findings, a count of external efforts cannot show that the vulnerable configuration received 15 audits, or any specific number.
Earlier warnings remain unresolved
James O’Beirne said he began examining Coldcard firmware in May 2025 to find out where its random-number generator got its entropy. In an August 4, 2026, account, he said he questioned the use of the libngu library, raised concerns about hardcoded constants in its software fallback, and recommended linking directly against libsecp256k1. He recalled being told the configuration was correct for physical devices and that any problem would already have become apparent. O’Beirne acknowledged that he did not pursue the issue rigorously.
On August 5, O’Beirne said he had received another credible account of a 2021 warning, separate from a Telegram discussion he had previously described. He suggested that as many as three people may have warned Coinkite. That possible number comes from his account and inference, not an independently established tally.
Available evidence does not show whether Dashjr’s findings, O’Beirne’s concerns, or the reported 2021 warning identified the same exploitable defect behind the theft. A private concern, a reproducible vulnerability report, and a completed audit are different things. Coinkite’s separate, detailed technical report was still in preparation in November 2026, leaving key questions about the review history unanswered.
What users with a potentially affected seed should do
Coinkite’s advisory says a firmware update cannot fix an existing seed generated under vulnerable conditions. Affected users need to create a new seed and move their funds. The reported migration instructions say to verify the new backup and receiving address, send a small test transaction, and transfer the remaining funds once the test succeeds.
Check Coinkite’s current security advisory for the precise affected-device and firmware thresholds. Available reporting does not provide a complete, independently verified list of affected model and firmware combinations. Users should not assume every Coldcard, or any particular model on its own, is affected. If you cannot confirm whether your seed is covered, follow Coinkite’s current guidance rather than relying on a general device list.
At the time of the November 2026 reporting, Coinkite’s security page listed standard releases 5.6.3 for Mk4/Mk5 and 1.5.3Q for Q. Firmware recommendations can change, so check the company’s live security page for current versions and instructions. Updating firmware and replacing an affected seed address different risks. A patched device does not make an old, vulnerable seed safe.
Key questions and answers
-
What caused the reported Coldcard vulnerability?
TRM Labs attributed it to a March 2021 build configuration error that selected a weak software random-number generator.
-
How much bitcoin was reportedly stolen?
TRM’s preliminary estimate, citing Galaxy Research, was about 1, 816 BTC from more than 5, 200 addresses across four waves beginning July 30, 2026. The roughly $116 million valuation reflected the bitcoin’s price at the time.
-
Does the “15 reviews” figure prove the faulty code was audited?
No. NVK said at least 15 external efforts examined Coldcard-related systems, but their full scope and findings are unavailable. The count does not establish how many examined the faulty build configuration.
-
Can a firmware update fix an affected seed?
No. Coinkite’s advisory says users with an affected seed need to create a new one and move their funds.
-
Has an attacker been identified or funds recovered?
Alex Thorn was reported as saying an identity “may be known to law enforcement.” Available information confirms neither an identification nor an arrest or recovery of funds.
The incident shows that hardware-wallet security depends on more than a device’s physical design or the number of past reviews. Seed generation, firmware builds, independent scrutiny, and clear migration guidance all matter. Self-custody gives users control, but it also demands that implementation failures be taken seriously. Bitcoin’s cryptography can remain intact while a wallet’s key-generation process lets users down. The Coldcard wallet theft shows why security depends on every step, not just the device itself. Self-custody has real risk when key-generation safeguards fail.
