OpenAI’s IH-Challenge Trains AI to Resist Prompt Injection
OpenAI’s IH-Challenge trains frontier LLMs to follow trusted instructions first, boosting safety and blocking prompt injection attacks. Here’s what it means.
AI Safety
Read More
GPT-5.4 Thinking System Card: What OpenAI Revealed
OpenAI released the GPT-5.4 Thinking system card on March 5, 2026. Here’s what the safety document actually says and why it matters right now.
GPT-5.4 Thinking System Card: What OpenAI Revealed
OpenAI published the GPT-5.4 Thinking system card. Here’s what it says about safety, reasoning limits, and what this model is actually allowed to do.
Google Defends Gemini With Medical Expert Safety Claims
Google has published a formal statement on the Gavalas lawsuit, arguing Gemini’s mental health safeguards are built alongside medical professionals.
Inside AI’s Black Box: How Mechanistic Interpretability Became 2026’s Biggest Research Breakthrough
For years, AI models have been treated as black boxes: data goes in, predictions come out, and nobody fully understands what happens in between. That’s changing. MIT Technology Review named mechanistic interpretability a top breakthrough technology for 2026, and the research coming out of Anthropic, OpenAI, and Google DeepMind is revealing how AI models actually…