OpenAI pauses training, Anthropic withholds model & more
OpenAI and Anthropic are pausing and withholding frontier models amid mounting cybersecurity risks and alignment concerns. Meanwhile, the EU begins enforcing its AI Act transparency rules as the FDA drafts guidelines for medical generative AI. Plus, Meta launches a desktop assistant with screen reasoning and Google DeepMind takes aim at deepfakes. Full edition: https://tech.pibrief.com/editions/2026-08-20
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.