CEOs, regulators flagged Anthropic model

Top United States officials and rival tech chiefs privately pressed Anthropic about a new artificial intelligence model before the company showed it to a wider group. (cnbc.com) The call took place about a week before Anthropic’s April 7 preview of Claude Mythos, according to CNBC and a Reuters follow-up. Vice President JD Vance and Treasury Secretary Scott Bessent were on the line with Anthropic chief executive Dario Amodei, Alphabet chief executive Sundar Pichai, OpenAI chief executive Sam Altman, Microsoft chief executive Satya Nadella, and the heads of CrowdStrike and Palo Alto Networks. (cnbc.com, yahoo.com) Anthropic did not put Mythos into general release. In its April 7 system card, the company said Mythos was its “most capable frontier model” and said the capability jump led it to limit access to a defensive cybersecurity program with selected partners instead. (anthropic.com) Anthropic says the worry is not chatbot misuse in the abstract, but software break-ins at machine speed. Its security team wrote that Mythos can identify and exploit previously unknown “zero-day” flaws in every major operating system and every major web browser they tested. (red.anthropic.com) A zero-day is a bug the vendor does not know about yet, which means there is no patch when an attacker finds it. Anthropic said more than 99% of the vulnerabilities Mythos found were still unpatched, so it withheld technical details and restricted access. (red.anthropic.com) The company’s answer is Project Glasswing, announced April 7. Anthropic said launch partners include Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, Microsoft, Nvidia, Palo Alto Networks and the Linux Foundation, with access also extended to more than 40 additional organizations. (anthropic.com) Anthropic said it is committing up to $100 million in usage credits and $4 million in donations to open-source security groups for that effort. The stated goal is to let defenders scan and harden critical code before attackers get similar tools. (anthropic.com) Concern spread beyond Silicon Valley to Wall Street this week. Reuters reported that Bessent and Federal Reserve Chair Jerome Powell convened an urgent meeting with bank chief executives in Washington to warn them about cyber risks tied to Anthropic’s model and similar systems. (finance.yahoo.com) That meeting included chiefs from Citigroup, Morgan Stanley, Bank of America, Wells Fargo and Goldman Sachs, according to Reuters’ account of the earlier Bloomberg report, while JPMorgan chief executive Jamie Dimon could not attend. Reuters said Goldman Sachs, Wells Fargo and the Federal Reserve declined to comment, and Treasury, lenders and Anthropic did not respond to its requests. (finance.yahoo.com) Anthropic has framed the government outreach as deliberate. The company told Nextgov that it briefed senior officials before any external release, including discussions with the Cybersecurity and Infrastructure Security Agency and the National Institute of Standards and Technology’s Center for AI Standards and Innovation. (nextgov.com) The immediate result is unusual: a frontier model was announced with a public system card, a restricted customer list and parallel briefings for the White House, regulators and banks. Anthropic’s line is that Mythos will stay limited while the company tests how to keep a tool built to find flaws from becoming one that helps exploit them at scale. (anthropic.com, cnbc.com)