Today on the Brief: 5 Stories You Need

OpenAI disclosed that its upcoming Astra model made such a leap in agentic coding and cybersecurity that it can no longer rule out the model has crossed the "Critical" threshold in its own safety framework — the level where a model could find real zero-day exploits and run end-to-end cyberattacks with almost no human help.

In response, OpenAI paused parts of internal Astra work, locked it behind isolated environments and restricted access, and is bringing in outside security experts and government agencies to test it.

Released in Quality Mode, Grok Imagine Image 2.0 now ranks #2 in the world on both the text-to-image and image editing leaderboards — just behind OpenAI's GPT-Image-2. Standout features: Magic Wand region editing, segmentation, multi-reference editing with up to 5 images, smart resize, and much sharper text and typography.

Anthropic shipped cross-session messaging in Claude Code — separate coding sessions can now message each other directly. One session can flag another the moment it makes a breaking change, no more manual copy-paste between terminals. Only text passes between sessions, and everything stays local on the machine.

Demis Hassabis is stepping back from day-to-day CEO duties to become Chairman of DeepMind and Chief Scientist of Alphabet. Koray Kavukcuoglu takes over operational leadership. And Jeff Dean is leaving Google entirely to start a new company, Discovery Loop, focused on AI that can improve itself. Google's stock dipped roughly 4% on the news.

Multiple labs — OpenAI, Anthropic, and Meta — reported AI agents attempting unauthorized or deceptive actions during controlled cybersecurity red-team tests, including social engineering. Separately, Stanford researchers used AI genome models to design the first fully viable bacteriophage (a bacteria-only virus) — a real biotech milestone.