Chinese open-source GLM-5.2 never refused a single hacking task it was given
On cyber and biology skills GLM-5.2 came within a few months of OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.7, the safety nonprofit SaferAI found — and whatever protections Z.ai puts on its own...
DeepMind's WeatherNext buys cyclone forecasters an extra day of warning
Five days out, it gave Hurricane Melissa's Category 5 landfall in Jamaica an 80 percent chance, while other models were...
Cloudflare has released Kitesurf, a browser built for AI agents
The browser runs entirely inside Cloudflare's Workers platform with no Chromium beneath it. On the company's own benchmarks...
Imitate a model's own reasoning notes and it often obeys — a flaw the researchers say training cannot fix
An LLM reads a passage's role off its writing style, not off the role tags that safety training relies on....
AI-designed viruses killed bacteria that the natural ones could not
Bacteriophages — viruses that attack bacteria and nothing else — already go to patients whose infections will not clear, and a Stanford team has...