OpenAI restricts internet access during frontier-model training as Anthropic confronts AI safety failures. Keeping AI under control is proving harder than promised.

More here: theregister.com/ai-and-ml/2026…
theregister.com/ai-and-ml/2026…