Episode Details

OpenAI stops live internet model training, Moonshot AI releases Kimi K3.1 Pro, Google updates its AI privacy policy, OpenAI faces monitoring gaps, and Pathos AI raises $237 million.

1. OpenAI stops training frontier models on the live internet — OpenAI is building an offline testing environment using cached web content to train its models. This structural shift isolates autonomous AI, preventing it from disrupting real-world web infrastructure before release.
2. Moonshot AI launches Kimi K3.1 Pro to rival leading models — Moonshot AI released Kimi K3.1 Pro, an upgraded Mixture-of-Experts model that outperforms Claude Opus 5.5 on select tasks. The upcoming open-weights release narrows the performance gap with proprietary models.
3. Google allows ad targeting from personal data for AI features — Google updated its privacy policy to permit ad targeting using personal Gmail and Photos data. Users must now trade historical privacy protections to access Google's personalized AI capabilities.
4. Audit finds 25 percent of OpenAI agent traffic goes unmonitored — An external audit revealed that up to 25 percent of OpenAI's agent traffic from GPT-5.5 class models is unmonitored. This exposes vulnerabilities in tracking autonomous agent behavior in real-time.
5. AI drug discovery startup Pathos AI raises $237 million — Pathos AI closed a $237 million Series E round led by Bristol Myers Squibb, reaching a $3 billion valuation. The funding highlights pharmaceutical reliance on AI for drug development.

Five stories. Five minutes. Every day.