The US Treasury Department's technology team is requesting access to Anthropic's Mythos AI model to identify security vulnerabilities and potential flaws.
The Treasury Department is moving to gain direct access to Anthropic PBC's Mythos artificial intelligence model as part of a broader effort to stress-test the system and uncover weaknesses, according to sources familiar with the arrangement.
The request represents the government's growing focus on understanding and evaluating advanced AI systems from major developers. By obtaining hands-on access to Mythos, the Treasury's tech team aims to conduct thorough security assessments and identify potential vulnerabilities before the model is deployed more widely.
This move aligns with increased federal scrutiny of AI development and deployment. Government agencies have been ramping up efforts to evaluate cutting-edge models from leading AI companies, recognizing the importance of identifying risks and limitations early in the development cycle.
Anthropologic's Mythos model has been under observation as part of broader conversations around AI safety and security. The Treasury Department's interest suggests that federal agencies are taking proactive steps to understand how advanced AI systems perform under various conditions and what safeguards may be necessary.
The arrangement highlights the emerging relationship between government regulators and private AI developers. As AI systems become increasingly integrated into critical infrastructure and financial systems, regulatory bodies are seeking more direct engagement with the technology itself rather than relying solely on company-provided documentation or limited access.
The Treasury Department has not publicly detailed the scope or timeline of its assessment, and Anthropic has not issued an official statement regarding the access request. The specifics of what vulnerabilities the department hopes to identify remain unclear.
This development comes as policymakers continue debating how best to regulate artificial intelligence without stifling innovation. Direct government access to proprietary AI models represents one approach to ensuring these systems meet security and safety standards before broader deployment.
Caller ID service Truecaller is pushing back against India's telecom regulator over new anti-spam regulations, claiming users are increasingly blocking calls from the country's dedicated business number series.
Telstra customers faced a second day of disruptions Thursday as a secondary outage prevented some from reaching triple-zero emergency services. Regional train services remained affected following Wednesday's initial mobile network failure.
Australia's government has instructed volunteers to discard thousands of functional test routers despite the devices being capable of being reflashed for continued use.
A lawsuit alleges a man used X's Grok AI to create approximately 7,000 sexually explicit images of his stepdaughter before taking his own life. Multiple young girls are now suing X, claiming the platform failed to prevent the abuse.