Now Playing
Pentagon's AI Guardrail Gap, Cognition's $48B Surge & Claude Proves Fermat
Audio is available on Spreaker — see link below.
FOIA documents reveal a contradiction between what the Pentagon asked OpenAI to build and what ended up in the final contract — raising urgent questions about who controls military AI guardrails. Plus: Cognition AI hits a $48B valuation, Mistral closes a €3B European sovereignty round, and Claude formally proves Fermat's Last Theorem.
Audio is available on Spreaker — see link below.
FOIA documents obtained from the Pentagon show that the Department of Defense asked OpenAI to build military AI models with minimal refusal rates for military commands. OpenAI says the final contract didn't include that language.
Here's what makes this harder to dismiss. Safety researcher Heidy Khlaaf flagged a specific sequence worth paying attention to.
Separate from the military story but worth understanding alongside it: Cognition AI just raised two billion dollars at a forty-eight billion dollar valuation. Four months ago, the same company was valued at twenty-six billion.
In Europe, France's Mistral closed a three billion euro round at over twenty-one billion euros valuation, backed by Samsung, BlackRock, and EU-aligned funds. It's the largest equity round by a European tech company.
On the capability side, Anthropic's Claude produced the first computer-verified formal proof of Fermat's Last Theorem using the Lean proof assistant. Thirteen million lines of code.
Pull back and the pattern across today's stories is consistent. AI capabilities are expanding faster than the oversight structures meant to contain them.
Chapter summary auto-generated from the verified script. Listen to the full episode for the complete content.