Anthropic’s Mythos AI model has drawn attention for its claimed ability to detect cybersecurity vulnerabilities, prompting concern among government agencies and institutions. Currently, only a select number of organizations have access to test the model internally. However, Jaya Baloo, COO and CISO at cybersecurity firm Aisle, challenges these claims based on her team’s own evaluations. She asserts that widely available open-source models, which are significantly less expensive, can identify the same types of bugs highlighted by Anthropic.
Baloo shared her findings during an interview with Caroline Hyde and Ed Ludlow on “Bloomberg Tech,” emphasizing that the performance gap between proprietary systems like Mythos and open alternatives may be overstated. Her insights suggest that the value of advanced AI in vulnerability detection might not be as unique as presented. The discussion underscores growing scrutiny over how AI capabilities are measured and marketed in the cybersecurity domain.