Thursday, April 9
AI Security Tensions Rise
Anthropic's decision to withhold its Claude Mythos (opens in a new tab) model from public release marks a watershed moment in AI safety policy. The model reportedly discovered thousands of zero-day vulnerabilities across major operating systems and browsers, prompting the company to launch Project Glasswing (opens in a new tab) with Apple, Microsoft, Google, and others for defensive cybersecurity applications only. Meanwhile, Chinese supercomputers suffered a massive breach with hackers stealing 10 petabytes (opens in a new tab) of data and attempting to sell the trove on underground markets, demonstrating the very real security risks that advanced AI capabilities could amplify if weaponized.
Meta launches its most ambitious AI push yet with Muse Spark (opens in a new tab), positioning the model as a step toward "personal superintelligence" and marking the company's return to serious competition with OpenAI and Google. The timing coincides with OpenAI facing renewed scrutiny after journalist Ronan Farrow (opens in a new tab) reported that a majority of over 100 interviewed sources described CEO Sam Altman as "a pathological liar," while separate reports suggest the company's internal models have solved five additional Erdős mathematical problems. The contrast between Meta's public confidence and OpenAI's leadership controversies signals a potential shift in AI industry dynamics.
Greece (opens in a new tab) becomes the latest country to restrict teen social media access with a ban for under-15s, citing anxiety and sleep problems, while Massachusetts (opens in a new tab) prepares to vote on similar legislation targeting both social media and cellphone use in schools. The regulatory momentum builds as Meta faces dual court losses (opens in a new tab) in the US, with legal observers noting that "accountability has arrived" for major tech platforms. These coordinated government actions across multiple jurisdictions suggest a fundamental shift away from the hands-off regulatory approach that allowed social media companies to operate largely unchecked for over a decade.