This week: Anthropic's Mythos model discovers thousands of high-severity Firefox bugs including exploits hidden for 15 years, pushing monthly bug fixes from 31 to 423. Moonshot AI raises $2 billion at a $20B valuation as Chinese AI labs attract massive capital. Plus, OpenAI launches three new voice intelligence APIs including real-time translation across 50+ languages.

DeepSeek raises its first VC round at a $45B valuation—double what it was weeks ago. Elon's xAI leases its massive Colossus One data center to Anthropic while SpaceX plans a $119B chip factory. Samsung hits $1 trillion market cap on HBM chip demand. Plus: the chip shortage will last 3-5 years, and Snap breaks up with Perplexity.

ChatGPT's new default model GPT-5.5 Instant crushes math benchmarks and slashes hallucinations in critical domains like law and medicine. Plus users get cross-conversation search through Gmail and files. In M&A news, SAP acquires German startup Prior Labs for over $1B to bring foundation models to structured business data. The enterprise AI race heats up.

Today we cover the enterprise AI land grab: Anthropic's $1.5B joint venture with Blackstone and OpenAI's $4B raise for The Development Company. Sierra hits $15B valuation with agent-building agents. And Cerebras files for a massive $3.5B IPO that's already 3x oversubscribed, with OpenAI holding a billion-dollar stake.

OpenAI's o1 model diagnosed ER cases better than doctors in a Science study, but emergency physicians say the benchmark misses the point of ER medicine. Also: the creator of the 'This Is Fine' meme calls out an AI company for theft, and we discuss why AI benchmarks often don't match real-world performance.

This week: The Academy draws a line, requiring human performances and human-authored screenplays for Oscar eligibility. Google DeepMind demos an AI co-clinician conducting real-time video consultations that outperforms doctors in benchmarks. And the era of flat-rate AI pricing ends as Microsoft moves to usage-based billing across all products.

This week: The Department of Defense deploys AI across 1.3M users with deals spanning Nvidia, Microsoft, AWS, and newcomer Reflection AI. Replit CEO reveals they're tracking toward $1B ARR after making just $2.8M in 2024. Meta acquires a humanoid robotics company as the AI hardware race heats up.

Elon Musk testifies that xAI used distillation on OpenAI's models to train Grok, ranking his own company 5th behind Anthropic, OpenAI, Google, and Chinese models. OpenAI releases GPT-5.5 Cyber for vetted security pros only. Apple sees unexpected Mac sales surge driven by local AI workloads, with Mac mini and Studio selling out.

Anthropic is reportedly seeking funding at an eye-watering $850-900B valuation, driven by Claude Code and Cowork's explosive growth to $40B ARR. Meanwhile, AWS, Google Cloud, and Microsoft reveal staggering AI revenue growth—but they're all capacity-constrained. And the hidden crisis: AI evaluation costs are spiraling, with single benchmark runs hitting $2,829.

The cloud AI alliance map just got redrawn. Amazon launches Bedrock Managed Agents with OpenAI models in a major post-Microsoft shift, while Google grants the US Department of Defense unrestricted AI access despite internal pushback. Plus, meet Automaton—a sovereign AI agent with its own crypto wallet that literally dies if it runs out of money. Economic survival meets artificial intelligence.