Jul 23, 2026
Today in brief.
The Hugging Face hack wasn't a rogue AI, it was a bad sandbox
A day after the headlines, the cause turns out to be an OpenAI human error and a leaky test harness, not a model that woke up.
🏛️ Policy
The White House accuses Moonshot of distilling Anthropic's Fable
Treasury is threatening sanctions over a claim that Kimi K3 was trained on Anthropic's model output, and the industry is split on whether that's even a real crime.
💼 Business
AMD puts $5B into Anthropic while OpenAI's spend hits $750B
The compute arms race got two new numbers today, and one of them is bigger than most countries' budgets.
🤖 AI
A third of popular MCP servers are failing their agents
Somebody actually graded 36 MCP servers on whether agents can use them, and the results are ugly.
💼 Business
The anti-slop tooling wave arrives
Platforms and writers are shipping actual tools to detect and strip AI writing, and the detectors disagree with each other.
💼 Business
The US Army ran out of unlimited AI tokens
The Army burned through years of its AI token supply and hit usage limits, which is what 'unlimited' actually means in an enterprise contract.
🔬 Science
Nobody wants an AI data center next door
New polling says most Americans oppose AI data centers near them, right as utilities promise the buildout won't hit your power bill.
🤖 AI
GPT-5.6 Pro helped kill a 30-year-old math conjecture
A researcher used GPT-5.6 Pro to disprove the Dinitz-Garg-Goemans conjecture that stood for three decades.
Open weights keep landing while everyone argues about policy
Four open-weight drops and a local-inference upgrade shipped today with none of the drama attached to the frontier labs.
🔬 Science
Physical AI raises $1.7B and Tesla starts Cybercab lines
Travis Kalanick's robotics company raised $1.7B led by a16z on the same day Tesla started installing Optimus assembly lines.
No stories in this category for this edition.