US-based Bitcoin security researchers, led by AnchorWatch CEO Rob Hamilton, are migrating to open-source Chinese AI models like Moonshot AI's Kimi K3 after facing repeated access revocations from OpenAI and Anthropic. The restrictions hamper the volunteer Bitcoin Red Team, which formed after a Coldcard wallet exploit leaked over $100 million. While American models block defensive audits, Chinese hackers are actively leveraging open-source AI agents to execute near-autonomous cyberattacks against Taiwanese government networks.
US AI cybersecurity restrictions
- ▪OpenAI and Anthropic have restricted US-based security researchers from using their frontier AI models for defensive cybersecurity audits of open-source Bitcoin repositories.
- ▪The Bitcoin Policy Institute published an open letter signed by over 70 organizations on August 10, 2026, calling on frontier AI labs to establish trusted-access programs for digital-asset defenders.
Bitcoin Red Team findings
- ▪During the first 30 hours of the audit, the Bitcoin Red Team scanned over 390 repositories, identifying 85 critical and 635 high-severity vulnerabilities with a 21% independent reproduction rate.
- ▪By August 8, 2026, the Bitcoin Red Team had scanned 501 projects, producing 7,958 findings, of which 1,280 were rated as high or critical severity.
- ▪The volunteer Bitcoin Red Team was formed to scan open-source Bitcoin repositories following a late-July 2026 firmware vulnerability in Coldcard hardware wallets that resulted in the theft of over $100 million in Bitcoin.
Chinese AI model adoption
- ▪AnchorWatch CEO Rob Hamilton and other US-based Bitcoin security researchers migrated their defensive auditing workflows to Kimi K3, an open-source model developed by Chinese AI company Moonshot AI.
- ▪Bull Bitcoin founder Francis Pouliot reported that an open-source Chinese model successfully identified, demonstrated on regtest, and helped patch a money-stealing exploit that American models refused to review.
OpenAI access revocations
- ▪On August 9, 2026, OpenAI's trust cyber program blocked AnchorWatch CEO Rob Hamilton from conducting security analysis on a Bitcoin codebase, despite Hamilton having completed onboarding and KYC processes.
- ▪Rob Hamilton was blocked again within 19 minutes of gaining access to OpenAI's Daybreak Blue cyber model while red-teaming Bitcoin infrastructure.
Autonomous AI cyberattacks
- ▪The AI-driven cyberattack on Taiwan cracked 85 government accounts within an hour, stole over 2,564 personnel records, and scanned nuclear safety and energy-sector networks within four days.
- ▪Chinese hackers used autonomous open-source AI agents, including Hermes and OpenClaw, to launch a near-autonomous cyberattack on Taiwanese government systems.
Story comments
Loading comments…