Multiple figures in the Bitcoin community and open-source developers have pointed out that AI models from China outperform their American counterparts on cybersecurity tasks, due to the strict usage restrictions imposed on US AI systems — restrictions so severe that some researchers have turned to Chinese models to help find vulnerabilities and secure critical Bitcoin infrastructure.
According to Bitcoin Magazine, Rob Hamilton, CEO of AnchorWatch — a Bitcoin self-custody insurance company — revealed that the restrictions on American AI models are obstructing his work. He recounted that after joining OpenAI's Trusted Cyber programme, through which he had completed identity verification and authentication months earlier, the system still refused to let him conduct further code analysis on a codebase he had already reviewed and disclosed vulnerabilities from.
Just days after Hamilton, one of the co-founders of Bitcoin Red Team, disclosed the problem and hinted that he might have to return to Chinese open-source models for research and to protect Bitcoin infrastructure, he was granted access to OpenAI's cybersecurity model Daybreak Blue — only to be restricted again after just 19 minutes of simulating attacks to test the security of systems against Bitcoin infrastructure.
Francis Pouliot, founder of Bull Bitcoin — a Bitcoin marketplace focused on self-custody — said he had never seen OpenAI this restricted before, and that it was not limited to security matters but extended to various Bitcoin-related topics as well. He warned that the American AI industry is finished if it continues down this path.
Pouliot added that he is fond of the Chinese open-source LLM Kimi K3, which genuinely helps with Bitcoin-related work.
**Bitcoin in serious trouble**
Meanwhile, a developer who goes by the pseudonym Calle, co-founder and lead software developer of Bitcoin Red Team, revealed that the team is using Chinese AI models to audit and find security vulnerabilities across virtually the entire open-source Bitcoin ecosystem.
Bitcoin Red Team was formed after Coldcard — long regarded as one of the most secure hardware wallets for self-custody of Bitcoin — was attacked and more than $100 million worth of Bitcoin was stolen in late July. This group of volunteers combines AI tools with human review to analyse wallets, Lightning applications, software libraries, and other Bitcoin-related projects. When credible vulnerabilities are found, they are reported privately to developers so issues can be fixed before details are disclosed publicly.
On Thursday (13 July), Calle posted on X that the world is facing a massive clash between open-source code accumulated by humans over decades and just two weeks of using Kimi K3, adding that everything is broken and Bitcoin is in serious trouble.
Kimi K3 is an AI model from Chinese startup Moonshot AI, which developers can download, install, and run on their own systems. The model is capable of analysing large codebases and handling complex, time-consuming software tasks with minimal human oversight.
The Red Team also uses the GLM 5.2 model from Z.ai, likewise a Chinese company, as well as models from American firms OpenAI and Anthropic.
However, the American models are more restrictive, and developers frequently encounter the limitations imposed by OpenAI and Anthropic when conducting security research.
Earlier this week, Calle posted that the Red Team had been blocked by OpenAI again, and complained about being tired of having to constantly ask for permission, ending the post by saying he was loading up Kimi K3.
He further noted that, despite the slow and difficult process, the Red Team has already been able to detect nearly all the easily discoverable issues and vulnerabilities in Bitcoin's open-source ecosystem.
This month, the Red Team disclosed that it has submitted 4,962 issue reports across 390 projects, of which 85 are classified as critical and 635 as high severity.
Calle added that response speeds vary greatly between projects, which reflects the status and quality of those projects. He also noted that Lightning-related software — designed to make Bitcoin payments faster and cheaper — is among the groups that are particularly difficult to audit due to its high complexity.
Calle further noted that projects that began AI audits several months ago are in a completely different state from those that have not yet done so. He recommended that each project will need its own AI auditing system going forward, and warned against trusting projects that are unmaintained or have ceased development.
Decrypt noted that Bitcoin Red Team is not the only one turning to Chinese AI. Last month, Hugging Face, an American AI startup, used China's GLM 5.2 model to investigate an incident in which the company's systems were breached by an OpenAI model, after American commercial models refused requests to analyse the attack incident logs.
It is worth noting that, despite saying Bitcoin is in serious trouble, Calle believes this major audit will make Bitcoin's software stronger in the long run. He concluded that Bitcoin is the first obvious target, but that before long other software systems around the world will have to face the same kind of AI-driven vulnerability scanning and auditing.
