AI Briefing

Anthropic accidentally leaked its most powerful model ever through an unsecured database - then confirmed it's real. Meanwhile, Claude is literally driving a Mars rover, Google just gave every US user a personal AI that reads their email, and Yann LeCun bet a billion dollars that LLMs are a dead end.

01ProductFortune ↗

Anthropic Accidentally Leaks 'Claude Mythos' - Its Most Powerful AI Model Yet

A Fortune exclusive revealed that Anthropic left details of an unreleased model called Claude Mythos in a publicly accessible, unsecured data store - along with nearly 3,000 unpublished blog assets. Rather than denying it, Anthropic confirmed the model exists and called it 'a step change' in capabilities, outperforming Opus in cybersecurity, coding, and academic reasoning. This is significant not just for the model itself but for the security implications - the company building AI it says poses 'unprecedented cybersecurity risks' leaked it through a misconfigured CMS.

02ResearchNASA JPL ↗

NASA's Perseverance Rover Completes First AI-Planned Drives on Mars Using Claude

NASA's Perseverance rover completed the first drives on another planet ever planned by artificial intelligence, covering 456 meters across two drives using Anthropic's Claude vision-language models. The AI analyzed orbital imagery and terrain data to plot waypoints - a task human operators had performed manually for 28 years. JPL verified over 500,000 telemetry variables through a digital twin before sending the commands to Mars. This marks a turning point for space exploration, where AI isn't just assisting but actively making navigation decisions on other worlds.

03Product9to5Google ↗

Google Rolls Out Gemini Personal Intelligence to All Free US Users

Google removed the paywall from Personal Intelligence, its feature that lets Gemini access your Gmail, Photos, Calendar, Drive, Maps, and YouTube data to provide personalized AI assistance. Previously restricted to paying AI Pro and Ultra subscribers, the feature now works across Google Search, Chrome, and the Gemini app for every US user with a personal account. This matters because Google is leveraging its unmatched data advantage - no other AI company has this level of access to users' personal information, making Gemini uniquely positioned as a true personal assistant.

04FundingTechCrunch ↗

Yann LeCun's AMI Labs Raises Record $1.03B Seed to Build World Models

Yann LeCun's startup AMI Labs raised $1.03 billion in what is believed to be the largest seed round in European startup history, at a $3.5 billion valuation. The Paris-based company is building 'world models' based on LeCun's Joint Embedding Predictive Architecture (JEPA), which he argues is a more promising path to machine intelligence than the autoregressive text prediction underlying ChatGPT, Claude, and Gemini. This is a massive bet against the current LLM paradigm by one of the most respected figures in AI, backed by names like Jeff Bezos, Eric Schmidt, and Mark Cuban.

05IndustryIBTimes ↗

OpenAI Hits $25B Revenue, Eyes Largest IPO in History at $1T Valuation

OpenAI has surpassed $25 billion in annualized revenue - reaching that milestone in roughly 39 months, faster than Salesforce (18 years), Google (17 years), or Facebook (12 years). CEO Sam Altman is steering toward what could be the largest IPO in stock market history, targeting a $1 trillion valuation with a potential filing in H2 2026. The catch: despite record revenues, OpenAI remains unprofitable and isn't expected to break even until 2030, with HSBC projecting a $207 billion funding shortfall through that date.