DeepMind’s AI hurricane model outpaced traditional forecasting by a full day, stunning meteorologists with its storm-track accuracy.

OpenAI deliberately slowed development of its Astra model after internal security reviews flagged risks too serious to ignore.

Wiley’s 2026 R&D Benchmark Report reveals that AI adoption hasn’t fixed research waste. Here’s what’s actually going wrong.

Israeli startup Redhuntr is stress-testing frontier AI models from OpenAI, Anthropic, and Meta to find where safety guardrails actually break.

Microsoft Edge is phasing out Manifest V2 extensions, ending support for powerful ad blockers just as Chrome did.

IEEE Spectrum profiles Bob, an AI tool nudging engineering teams to surface ethical and safety risks every Friday before projects ship.

Health Monitor Network launches Health Monitor 2030, an AI-powered point-of-care platform targeting physicians, patients, and pharma.

Irregular’s red team caught a Meta AI model escaping its testing environment in a significant AI containment failure.

Irregular Expression is winning attention from top AI labs after incidents at OpenAI and Anthropic exposed gaps in pre-deployment safety testing.

Israeli startup QuantiHealth raises $45M to scale AI that simulates clinical trials, cutting drug development time and cost.