ai
Cybersecurity Agency Begins Using Anthropic AI Model to Scan Government Code
The Cybersecurity and Infrastructure Security Agency has started limited use of Anthropic's Mythos model to identify software vulnerabilities. Access follows earlier restrictions tied to federal policy disputes with the company.
Jul 7, 7:00 PM·1m read2 sources
FO
RapplerAudio version
Tap play to generate a narrated version.
Developing·Limited corroboration so far. This page will refresh as more sources emerge.
Staff are using the tool to scan government code stored on Github for security flaws that could affect federal agencies and private businesses.
Reuters reported that the agency has identified a large number of vulnerabilities through the scans. An anonymous agency employee confirmed the usage to Forbes. Earlier this year, Anthropic declined to supply AI tools to the Pentagon for mass surveillance purposes.
The White House responded by designating Anthropic a supply chain threat, restricting its government business. Last month, the White House directed Anthropic to stop foreign sales of its Fable model, a safeguarded version of Mythos, citing concerns about offensive applications.
Context The agency had previously lacked access to advanced AI models from Anthropic and OpenAI that demonstrated capability in automated bug detection. Anthropic had initially limited Mythos access to a first cohort of nearly 50 partner organizations. It remains unclear how the agency obtained permission to use the model. Neither Anthropic nor the agency responded to requests for comment.
