He Pentagon It is clear: Anthropic continues to represent a supply chain riskbut Mythosthe artificial intelligence (AI) model developed by the company and with advanced cyber capabilities, is a «separate issue for national security». This was stated by the director of technology of the Department of Defense, Emil Michael, during an interview on ‘CNBC’, in which he stated that «the Mythos issue is being addressed at the government level«Not only in the War Department, it is a separate national security issue in which we have to ensure that our networks are reinforced, because that model has specific capabilities to find cyber vulnerabilities and correct them.» These statements come at a time of tension between the Pentagon and the company, which refuses to allow military to use AI chatbot Claude for US mass surveillance or the deployment of autonomous weapons due to safety and ethics concerns. The White House has already informed Anthropic that opposes its plan to expand access to this model to 70 additional organizationswhich would bring the total number of companies that can use it to around 120 entities. In this sense, he explained that Mythos is powerful enough to detect and exploit vulnerabilities in critical software, which makes it an offensive tool that can be used against the United States. However, it is worth remembering that Anthropic has initially limited its launch to a few major technology and finance companies. Those companies, which include Nvidia, OpenAI, Google, Microsoft, amazon and Appleare part of the call «Project Glasswing»which will work to protect critical systems before other similar AI models become available. On the other hand, the government also believes that does not have enough processing capacity for 120 clients to use Claude Mythos without degrading the performance of the model in government agencies that already use it. Mythos is a more powerful AI model capable of identify and exploit weaknesses in «all major operating systems and all major web browsers» if a user orders it to do so. What’s more, Anthropic claims that AI models «have reached a level of programmability where they can outperform all but the most skilled humans at finding and exploiting software vulnerabilities.»
