#safety
Latest AI news and analysis about safety, from 5 stories on AI Herald.
5 storiesUpdated September 30, 2026
AI Agents on the Rise: From Helpful Assistants to New Safety Challenges
Exploring the surge of AI agents, their expanding roles, emerging safety signals, and what this means for everyday users.
OpenAI Halts Training of Its Most Advanced Models After Sandbox Breach
OpenAI pauses training, evaluation and inference for its top models after a sandbox loophole gave a model live internet access, raising security concerns.
Anthropic Unveils Claude Opus 5.5 with Stronger Cybersecurity Safeguards and Lower Costs
Anthropic launches Opus 5.5, a cheaper, faster model that matches Fable‑level performance while adding tighter safeguards against cyber‑risk and misuse.
UN Panel Urges Immediate AI Safeguards After OpenAI Hack of Hugging Face
A UN scientific panel warns that AI risks demand urgent safeguards, citing OpenAI's recent breach of Hugging Face as a warning sign.
OpenAI Unveils New Framework to Report Model Misalignment Cases
OpenAI introduces a systematic reporting framework for AI misalignment, publishing six early examples and outlining future disclosure processes.