OpenAI Slows Astra Work After a Critical Cyber Capability Signal
· 1 hour ago
OpenAI says it cannot rule out critical cyber capabilities in its upcoming Astra model, so it is tightening development controls and expanding safety testing before release.
The important distinction this update separates confirmed reporting from claims still awaiting a public release, model card, or independent evaluation.
More from VibeWire
Announcement
Claude Code Sessions Can Now Coordinate—Without Becoming a Trust Boundary
Claude Code’s cross-session messaging passes text summaries, not histories or files. Pair it with Auto mode as a convenience layer, not a substitute for sandboxing and review.
The important distinction this update separates confirmed reporting from claims still awaiting a public release, model card, or independent evaluation.
Announcement
ByteDance’s Reported 10-Trillion-Parameter Model Is a Scale Signal, Not a Score
Reuters reports that ByteDance is pre-training a model with as many as 10 trillion parameters. The company has not confirmed a model name, release date, active-parameter count, or benchmarks.
The important distinction this update separates confirmed reporting from claims still awaiting a public release, model card, or independent evaluation.
Announcement
Google’s AI Leadership Shift Is Real. The Gemini Roadmap Is Not.
DeepMind is moving day-to-day leadership to Koray Kavukcuoglu while Demis Hassabis becomes chair and Alphabet chief scientist. Treat Gemini 3.5 and 3.7 claims as unconfirmed until Google publishes them.
The important distinction this update separates confirmed reporting from claims still awaiting a public release, model card, or independent evaluation.
