Bitcoin Red Team returns to Chinese AI models after OpenAI block
An incident that exposed a serious problem of Western cyber defenders' dependence on corporate AI platforms occurred in early August. AnchorWatch CEO Rob Hamilton, an active participant in the volunteer initiative Bitcoin Red Team, reported that on August 8 he began integrating OpenAI's cyber capabilities into his work. However, by the next morning, access to the service was blocked, despite the fact that he had previously completed the full KYC verification and onboarding procedure.
Hamilton emphasizes that the access restriction occurred in the midst of an audit of code changes, which prevented him from completing the review for potential vulnerabilities. In his view, such a policy by American AI companies puts defenders at a deliberately disadvantageous position, cutting them off from the most powerful tools, while attackers are not bound by any corporate restrictions.
When asked about Anthropic's policy regarding trusted access to the most powerful models, he noted that no problems have arisen so far, but suggested that he might be downgraded from the Fable level to Opus, which is essentially the same problem.
Key figures and context
Bitcoin Red Team is an independent group that uses AI and manual review to analyze hundreds of open-source repositories related to bitcoin. As of August 9, a team of 25 participants had found 1,288 critical and high-level vulnerabilities in the bitcoin ecosystem. This is a colossal amount of work that underscores the critical importance of access to advanced AI tools.
Notably, despite the block, the team reports that OpenAI and Anthropic cyber models deliver "stunning results." However, as Hamilton ironically noted, a "very Chinese period of life" is beginning. Total AI expenses have already exceeded $58,000, with nearly 75% of all costs attributed to the Chinese model Kimi K3, and the share of Qwen 3.8 is also growing rapidly.
This case is a vivid example of how geopolitical restrictions in the AI sphere hit Western initiatives themselves. While 25 American companies in July called on authorities not to restrict Chinese models with open weights, OpenAI's actions show the opposite trend. In my view, this will only accelerate the transition of serious research groups to open and decentralized AI solutions that do not depend on the whims of a single corporation. The irony is that the attempt to restrict Chinese technology leads to its even more active adoption in critically important sectors, including cybersecurity.