Bitcoin Red Team's AI Audit Found 85 Critical Vulnerabilities — Here's What Happened After the $116M Coldcard Hack

How It Responded to the Coldcard Wallet Hack

The campaign was a direct response to the July 2026 Coldcard hardware wallet exploit . Here is what happened with that hack:

• The vulnerability: A firmware bug introduced in Coldcard firmware version 4.0.0 (March 2021) caused the device to bypass its dedicated hardware random-number generator during seed-phrase creation . This made wallet seed phrases predictable — key strength dropped from 128 bits to as little as 40 bits .
• The attack: Beginning July 30, 2026, attackers exploited this flaw to drain funds from thousands of wallets. The first wave hit 1,196 addresses in just 41 minutes, stealing 1,082.65 BTC (~$70M at the time) .
• Total losses: Multiple waves pushed the total past $116 million across over 5,200 addresses, with some estimates exceeding $130 million as the attack continued . At least a dozen different hackers were involved .
• Coinkite's response: The company urged users to update firmware and move funds, but by then much of the damage was done .

In response, the Bitcoin Red Team launched its audit to find similar random-number-generator bugs, weak cryptographic practices, and other systemic flaws before attackers could exploit them .

Key Findings and Methods

Findings

The sheer volume — nearly 5,000 reports in a day — created a major triage problem for downstream maintainers . Only about one in five findings was confirmed reproducible, but the group argued that flagging borderline issues was justified given the severity of the Coldcard attack .

Methods

• The team used open-weight AI models to scan codebases for known vulnerability patterns, cryptographic weaknesses, and logic errors .
• Human researchers guided the AI, validated results, and filed responsible-disclosure reports .
• The group plans to open-source its AI security harness so the broader community can continue automated scanning .
• They focused on wallets, cryptographic libraries, and Bitcoin infrastructure projects — the kinds of code that, if flawed, could lead to another Coldcard-scale loss .

Why OpenAI Revoked the Lead Researcher's Access Mid-Project

Rob Hamilton (AnchorWatch CEO and campaign co-lead) had enrolled in OpenAI's "Trust Cyber" program, which grants approved security researchers subsidized API access for vulnerability research . On August 8/9, 2026, after he had already integrated the program into his workflow and filed valid disclosures, OpenAI revoked his access without explanation .

• Hamilton publicly stated he was "blocked from doing additional analysis on a codebase which I've already responsibly disclosed to, and have received confirmation had legitimate findings" .
• The revocation forced the Bitcoin Red Team to pivot to less restrictive AI models, including Chinese open-weight alternatives, a shift that was widely noted and criticized as a setback for U.S. AI competitiveness in security research .
• OpenAI has not publicly detailed its reasoning. The company's Researcher Access Program FAQ notes credits are granted at their discretion and can be withdrawn, but no specific policy violation by Hamilton has been cited .