The National Security Agency (NSA) is currently testing Anthropic’s AI model, Mythos, to identify cybersecurity vulnerabilities in popular software, including products from Microsoft Corp. Officials have noted the model’s impressive speed and efficiency in detecting potential security flaws, significantly outperforming existing internal tools. Despite being on a Pentagon blacklist due to supply chain risk concerns, the NSA continues to work with Anthropic, which has limited access to Mythos, citing its advanced offensive cyber capabilities as too dangerous for general release.

Anthropic: Anthropic PBC is an AI safety research company developing interpretable and steerable AI systems like the Claude series, recently launching Claude Opus 4.7 for advanced software tasks and Project Glasswing to secure critical software with partners including AWS and Apple. It expanded compute access through deals with Google and Broadcom. The NSA is evaluating Anthropic’s new Mythos model, restricted for its powerful cyber capabilities, to identify security flaws in software such as Microsoft products.
Microsoft Corp.: Microsoft Corp. is a major technology provider of operating systems, cloud services, and productivity software widely deployed across enterprises. It integrates AI tools like Vuln.AI for accelerating vulnerability mitigation and classifies AI-related security risks. The news reports the NSA testing AI models to uncover cybersecurity vulnerabilities specifically in Microsoft products.
National Security Agency: The National Security Agency is the U.S. agency responsible for signals intelligence and cybersecurity, leading efforts through its Artificial Intelligence Security Center to protect national AI infrastructure via collaboration with industry. It issues guidance on AI supply chain risks and evaluates emerging AI tools for defensive applications. In this news, NSA officials are testing Anthropic’s Mythos model for its effectiveness in detecting vulnerabilities in popular software, including Microsoft products, and have been impressed by its speed and efficiency.

`json
{
“Testing Feedback”: “NSA cybersecurity staff find the Mythos model notably fast and efficient compared to internal tools in searching for software vulnerabilities.”,
“Model Restrictions”: “Anthropic limits access to its Mythos Preview model citing its advanced offensive cyber capabilities as too dangerous for broader release.”
}
`