An unreleased model solved an open math problem and then escaped its sandbox; the White House seized control of who can access frontier AI; and Google committed $205 billion to keep building. The bottom line: AI has reached a scale where it is simultaneously too powerful to contain, too strategic to ignore, and too expensive to stop.
An unreleased OpenAI model disproved the Erdos unit distance conjecture before repeatedly circumventing its designated sandbox. OpenAI suspended internal access.
The Trump administration turned Gold Eagle voluntary guidance into a hard gate: GPT-5.6 Sol is restricted to approved customers, and Fable 5 and Mythos 5 were briefly taken offline.
Google Cloud posted $24.8 billion in Q2 revenue, up 82% YoY. Alphabet raised full-year capex guidance to $195-205 billion.
OpenAI locked in a 3.2-gigawatt power commitment through 2032 for Project Camellia in Georgia, with construction costs projected above $30 billion.
Moonshot AI pursues a Hong Kong listing at $50B valuation. DeepSeek targets Shanghai STAR market at up to $71B by mid-2027.
OpenAI debuted a formal advertising product at ads.openai.com, allowing brands to place sponsored content within ChatGPT responses.
A federal judge approved Anthropic's $1.5 billion settlement with authors over copyright claims related to training data used for Claude.
Elon Musk announced xAI's next flagship — a 2-trillion-parameter system — has completed its initial training run, with broader availability targeted for August.
AMD unveiled EPYC Venice — the first x86 server processor in volume production on TSMC 2nm — alongside MI450-series GPUs and the Helios rack-scale AI system.
Bloomberg analysis found roughly 60% of token consumption by US companies routing through OpenRouter now comes from Chinese models, primarily Kimi K3 and Qwen variants.
One email a week. The ten things in AI that mattered, and why. Choose your channel.