CEOs, govt discuss Anthropic Mythos
Executives from Google, OpenAI, Microsoft, CrowdStrike and others joined a conference call with the U.S. government over safety fears about Anthropic’s new model ‘Mythos’. The meeting took place ahead of a wider preview of the system, signalling government-level scrutiny of frontier models. (timesofindia.indiatimes.com)
U.S. officials held a private call with top tech chiefs before Anthropic widened access to Claude Mythos Preview, a model Anthropic says is unusually strong at cybersecurity work. (cnbc.com) The April 2026 call included Vice President JD Vance, Treasury Secretary Scott Bessent, Anthropic chief executive Dario Amodei, OpenAI chief executive Sam Altman, Google chief executive Sundar Pichai, Microsoft chief executive Satya Nadella, CrowdStrike chief executive George Kurtz and Palo Alto Networks chief executive Nikesh Arora, according to CNBC and Times of India’s account of that report. (cnbc.com, timesofindia.indiatimes.com) Anthropic announced Claude Mythos Preview on April 7, 2026 and said it would not make the model generally available. Instead, it limited access to a defensive security program called Project Glasswing. (anthropic.com, cnbc.com) Large language models are systems trained to predict the next token in text, code and other data, which lets them generate answers and write software. Anthropic says Mythos stands out because it can find software flaws and help turn them into working exploits, the code attackers use to break into systems. (anthropic.com, red.anthropic.com) Anthropic’s security team wrote that Mythos was capable, in testing, of identifying and exploiting zero-day vulnerabilities in every major operating system and every major web browser. A zero-day is a previously unknown bug, which means defenders have had zero days to patch it before discovery. (red.anthropic.com) Anthropic said more than 40 organizations received access for defensive work, and it named Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA and Palo Alto Networks as launch partners in Project Glasswing. Anthropic also said it committed up to $100 million in usage credits and $4 million in donations to open-source security groups. (anthropic.com) That limited rollout came after Anthropic said Mythos showed a “striking leap” over Claude Opus 4.6 on many benchmarks and that the increase in capability led the company to keep the model out of general release. Anthropic’s system card says the model is its “most capable frontier model to date.” (anthropic.com) CNBC reported that the White House call focused on two questions: how secure the models themselves are, and how the United States should respond if artificial intelligence starts giving attackers an edge over defenders. Reuters, in a pickup of CNBC’s reporting, said officials raised the issue about a week before Anthropic released Mythos. (cnbc.com, msn.com) The government outreach did not stop with Silicon Valley. TechXplore and CNBC reported that Bessent and Federal Reserve Chair Jerome Powell also met bank chiefs that same week to discuss the security implications of Mythos for the financial system. (techxplore.com, cnbc.com) Anthropic’s public position is that the same capability can help defenders move faster if access is tightly controlled. In announcing Project Glasswing, the company said it wanted to give infrastructure owners “a head start” before tools like Mythos become more widely available across the industry. (anthropic.com, cnbc.com) For now, the clearest fact is the release decision: Anthropic built a stronger model, kept it behind a narrow gate and triggered direct calls with Washington before a broad public launch. (anthropic.com, cnbc.com)
