The Bitcoin Red Team has turned to Chinese AI models after facing access restrictions from OpenAI, raising concerns about cybersecurity tools for defenders.
Rob Hamilton, the CEO of AnchorWatch, revealed that he began incorporating OpenAI's technology into his cybersecurity operations with the Bitcoin Red Team on August 8. However, he faced access issues the following day, prompting him to switch to Chinese AI models instead.
Hamilton clarified that these restrictions came despite having completed the necessary KYC and onboarding months earlier to utilize OpenAI's cybersecurity tools.
Due to these limitations, Hamilton expressed frustration over his inability to audit code changes or evaluate potential vulnerabilities effectively. He raised concerns that American AI firms' policies hinder defenders while leaving malicious actors unaffected.
In response to inquiries about Anthropic's access policies, Hamilton noted he has yet to encounter issues but anticipated a downgrade from Fable to Opus, which he described as a similar predicament.
The Bitcoin Red Team, comprising 25 volunteers, employs both AI and manual checks to scrutinize numerous open-source Bitcoin repositories. As of August 9, they had identified 1,288 critical and high-level vulnerabilities within the Bitcoin ecosystem.
On Twitter, Hamilton remarked that he and fellow participant calle were entering a "very Chinese period of life" due to the shift in AI models.
The team acknowledged that while AI models from OpenAI and Anthropic deliver impressive results, nearly 75% of their AI expenses still go towards Kimi K3, with an increasing share for Qwen 3.8. They reported a cumulative expenditure exceeding $58,000 as they diversify their inference sources.
Despite the challenges, our AI spending remains high, and we're seeing remarkable outcomes from the various models we're integrating.
Notably, a group of 25 U.S. companies recently urged the government to avoid imposing restrictions on Chinese AI models, highlighting the ongoing debate surrounding AI access and security.
