Bitcoin security researchers say the restrictions on American frontier AI models from OpenAI and Anthropic have become so severe that they are now relying on Chinese open-source models to defend critical infrastructure. The warning, documented by Bitcoin Magazine — which also distributes its reporting through its official news app — comes from executives and core developers across the digital-asset ecosystem, who describe U.S. models refusing or blocking legitimate defensive security work even for users granted explicit access Bitcoin Magazine.
Rob Hamilton, CEO of Bitcoin self-custody insurer AnchorWatch, said he completed know-your-customer verification for OpenAI’s trusted cyber program months earlier, yet was blocked from analyzing a codebase he had already responsibly disclosed. Days later, after gaining access to OpenAI’s “Daybreak Blue” cyber model, he was blocked again within 19 minutes while red-teaming Bitcoin infrastructure, according to the report Bitcoin Magazine.
Francis Pouliot, founder of Bull Bitcoin, said a Chinese open-source model identified a money-stealing exploit in a project he was auditing, demonstrated it on a regtest network, and helped patch it — while the American models he pays for refused to review the same fix. The contrast has pushed developers toward open-weight Chinese models such as Kimi K3, the report notes Bitcoin Magazine.
The friction centers on OpenAI’s purpose-built security models. The company recently made its Daybreak AI security models available through AWS Bedrock, a move covered separately on zBrandco OpenAI adds Daybreak AI security models to AWS Bedrock, yet researchers say access for independent defenders remains inconsistent.
A volunteer effort called the Bitcoin Red Team, led by Cashu creator Calle and Hamilton, used Kimi K3 as its primary engine to audit hundreds of open-source repositories after a Coldcard firmware flaw exploited from July 30 stole more than $100 million in bitcoin. By August 8, after 100-plus hours of work, the group said it scanned 501 projects and produced 7,958 findings, of which 1,280 were rated high or critical severity, with most compute spent on Chinese open-weight models Bitcoin Magazine.
On August 10, the Bitcoin Policy Institute published an open letter signed by more than 70 organizations — custodians, exchanges, miners, and developer groups — demanding frontier labs create trusted-access programs for qualified defenders, including early model access, compute, secure review environments, and direct security-team channels, as detailed in the report Bitcoin Magazine.
Calle argues the human-only era of open-source review is over and urges every project to run a permanent AI audit pipeline, while warning developers to stop writing security-critical code in C because memory-safety bugs that once needed a skilled hacker to exploit are now “a single prompt.” The report frames Bitcoin as the first major open-source ecosystem forced to confront frontier AI’s collision with accumulated human code Bitcoin Magazine.
