Anthropic's Secret 'Claude Mythos' Model Exposed in Embarrassing Security Leak
In what might be the most ironic AI story of the year, Anthropic accidentally leaked details of its most powerful model yet through an unsecured, publicly accessible data store. Security researchers from LayerX Security and the University of Cambridge discovered close to 3,000 unpublished assets — including draft blog posts describing a new model tier called Claude Mythos (also referred to internally as "Capybara").
An Anthropic spokesperson confirmed the model represents "a step change" in AI capabilities and is "the most capable we've built to date." According to the leaked drafts, Mythos sits above the current flagship Opus line, delivering "dramatically higher scores on tests of software coding, academic reasoning, and cybersecurity" compared to Claude Opus 4.6.
The cybersecurity angle is particularly striking: the draft blog post describes Mythos as "currently far ahead of any other AI model in cyber capabilities," while warning it "presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders." The irony of a company building unprecedented cybersecurity AI while leaving its own announcements in an unsecured database needs no elaboration.
Anthropic attributed the incident to "human error" in CMS configuration and said the rollout will begin with a small group of early-access customers focused on cybersecurity evaluation. The company acknowledged the model will be "very expensive" to serve.
Sources: Fortune, Futurism, The Decoder
OpenAI Pulls the Plug on Sora
OpenAI is shutting down Sora, the AI video generator that launched to massive fanfare in February 2024 and briefly topped Apple's App Store charts. The company confirmed the decision on March 24, saying simply: "We're saying goodbye to Sora."
The reasons are pragmatic. Sora's GPU-intensive video generation was straining OpenAI's infrastructure — the head of Sora said last November that the firm's GPUs were "melting." With an IPO potentially on the horizon and annualized revenue now exceeding $25 billion, OpenAI is reallocating compute resources toward more lucrative coding, reasoning, and text-generation workloads, as well as robotics research.
The shutdown also killed an unfinalized $1 billion Disney partnership that would have let users create videos with Disney's copyrighted characters. "We respect OpenAI's decision to exit the video generation business," a Disney spokesperson said. The wind-down will be phased: new features are already frozen, users get a transition window to download their content, and then the app and API go dark permanently.
OpenAI isn't fully exiting AI video — video generation remains available inside ChatGPT — but Sora as a standalone product is ending as a footnote rather than a game-changer, especially as competitors like Google Veo, Runway, and Kling caught up or surpassed it.
Shopify's Agentic Storefronts: Buy Anything, Anywhere AI Talks
E-commerce just entered the AI chat era. Shopify launched Agentic Storefronts for all eligible US merchants on March 24, enabling products to appear and be purchased directly through ChatGPT, Google's AI Mode, Microsoft Copilot, and the Gemini app. The feature is on by default — most Shopify stores are already discoverable in these AI platforms without merchants lifting a finger.
The numbers are staggering: 5.6 million merchants now have a distribution channel reaching 880 million monthly active ChatGPT users. AI-driven traffic to Shopify stores is up 7x since January 2025, with AI-attributed orders up 11x over the same period.
On the fee side, OpenAI takes a 4% cut on ChatGPT checkout transactions, while Google AI Mode and Gemini currently charge nothing. Shopify also announced the Universal Commerce Protocol (UCP), an open standard co-developed with Google and backed by Walmart, Target, Etsy, Stripe, Visa, and Mastercard — essentially trying to make AI-native commerce an industry standard rather than a platform-specific feature.
White House Unveils National AI Policy Framework
The Trump administration released its National Policy Framework for Artificial Intelligence on March 20, laying out a blueprint for Congress to create unified federal AI regulation. The framework's most consequential proposal: federal preemption of state AI laws, arguing that a patchwork of 38 states' AI legislation is stifling innovation.
Key provisions include no new federal AI regulator (existing agencies keep sector-specific oversight), child safety protections with age assurance requirements, and a copyright stance that training on copyrighted material doesn't violate copyright law — though the administration would leave courts to settle specific disputes.
The framework drew immediate criticism from AI watchdog groups. Americans for Responsible Innovation president Brad Carson said it "shields AI developers from liability," while Democratic lawmakers raised concerns about weakened accountability and oversight. The White House wants the framework codified into law this year.
Meanwhile, the EU Council agreed its position on streamlining the AI Act, proposing to extend timelines for high-risk AI system rules by up to 16 months and adding new prohibitions on AI-generated non-consensual intimate content.
Sources: CNBC, Nextgov, EU Council