AI Briefing

Anthropic just casually dropped the most dangerous AI model ever built, found thousands of zero-day vulnerabilities in every major OS and browser, and then said 'yeah, we're not releasing this one to the public.' Meanwhile Meta finally shipped Muse Spark and it's... fine? Competitive but not leading. And Anthropic quietly revealed it's tripled revenue to $30 billion and locked in 3.5 gigawatts of Google TPU capacity, because apparently the AI arms race now runs on power plant math.

01ProductTechCrunch ↗

Anthropic Launches Claude Mythos Preview via Project Glasswing, Finds Thousands of Zero-Day Vulnerabilities

Anthropic released a preview of Claude Mythos, its most capable model yet, but in a first for the industry chose not to make it publicly available due to its extraordinary cybersecurity capabilities. Through Project Glasswing, a limited partnership with 12 organizations including Amazon, Apple, Google, Microsoft, NVIDIA, and CrowdStrike, the model is being deployed exclusively for defensive security work. Mythos Preview discovered thousands of previously unknown zero-day vulnerabilities across every major operating system and web browser, autonomously constructed sophisticated exploit chains, and achieved 181 working Firefox JavaScript exploits compared to just two from earlier models. Anthropic is backing the initiative with $100 million in usage credits and $4 million in donations to open-source security organizations, pricing the model at $25/$125 per million input/output tokens for partners.

02ProductCNBC ↗

Meta Debuts Muse Spark, Its First Proprietary Model From Superintelligence Labs

Meta launched Muse Spark on April 8, the first model from its new Meta Superintelligence Labs division led by former Scale AI founder Alexandr Wang. The model is purpose-built to be lightweight and consumer-centric, accepting voice, text, and image inputs with a novel 'contemplating mode' that lets multiple AI agents collaborate on complex questions in science, math, and health. Muse Spark ranked fourth on the Artificial Analysis Intelligence Index with a score of 52, trailing Gemini 3.1 Pro (57), GPT-5.4 (57), and Claude Opus 4.6 (53). Notably, this is Meta's first closed proprietary model, a sharp departure from its open-source Llama strategy, though the company says it hopes to open-source future versions. Meta shares surged nearly 9% on the news.

03ProductOpenAI ↗

OpenAI Merges ChatGPT, Codex, and Atlas Into Unified 'Super App' Alongside ChatGPT 5.5

OpenAI launched a unified desktop application on April 6 that merges ChatGPT, its Codex coding agent, and the Atlas AI browser into a single interface, alongside the release of ChatGPT 5.5. The super app lets users switch between writing, coding, and browsing without leaving one window, targeting the kind of workflow integration that has made tools like VS Code sticky for developers. ChatGPT 5.5 is a bridge release with improved memory management across long sessions, better task continuity when switching between work streams, and early previews of GPT-6 capabilities. Available immediately to Plus and Pro subscribers with a free-tier rollout to follow, the move signals OpenAI's ambition to become the default operating layer for knowledge work, not just a chatbot.

04IndustryAnthropic ↗

Anthropic Revenue Triples to $30 Billion Run Rate, Signs Historic 3.5GW Compute Deal with Google and Broadcom

Anthropic revealed its annual run-rate revenue has surged to $30 billion, tripling from approximately $9 billion at the end of 2025, while simultaneously signing its largest infrastructure deal yet: 3.5 gigawatts of next-generation Google TPU capacity through Broadcom, coming online starting in 2027. This adds to the 1 gigawatt already being supplied in 2026. The company now has over 1,000 enterprise customers each spending more than $1 million annually, a figure that has more than doubled since February. Analysts at Mizuho estimate Broadcom will earn $21 billion in AI revenue from Anthropic in 2026 and $42 billion in 2027, underscoring just how massive the compute infrastructure buildout behind frontier AI has become.