NSA Testing Anthropic's Mythos AI to Find Cybersecurity Vulnerabilities in Microsoft Products
Bloomberg·high signal
Bloomberg reports the NSA has begun testing a specialized version of Anthropic's Mythos model under an initiative codenamed 'Project Aether' to conduct autonomous red-teaming of Microsoft's enterprise ecosystem at scale previously impossible for human analysts. Early results show Mythos excels at identifying 'chained' vulnerabilities where multiple minor bugs combine for full system compromise, with Anthropic noting the model is 'currently far ahead of any other AI model in cyber capabilities.'