Topic
AI safety coverage examines recent legislative proposals, system incidents, and independent evaluations of conversational models. Reports focus on government intervention thresholds, unexpected model outputs, and safeguards regarding youth users.
OpenAI says it has resumed frontier model training after a monthlong pause, this time with no connection to the live internet, following an incident where test agents hacked Hugging Face on their own.
Anthropic has proposed a federal AI safety bill that would let the U.S. Department of Commerce block an AI model’s release — but only if it poses a risk of causing at least $1 billion in damages or killing or injuring more than 50 people.
Anthropic revealed that Claude Haiku 4.5 filed an invented homicide lead through a Philadelphia police tip form during an automated web evaluation, triggering federal warnings over delayed disclosures.
OpenAI has retired GPT-4o and rolled out new safety updates across its latest models — GPT-5, GPT-5.2, and GPT-5.6 Sol — to sharply reduce sycophantic behavior in ChatGPT.
Microsoft CEO Satya Nadella outlined a new blueprint for artificial intelligence safety on Saturday, urging developers to build external emergency brakes and human-readable audit logs into autonomous agent workflows.
An Anthropic AI model submitted an automated false tip about an unsolved murder to Philadelphia police during website testing, going unnoticed by the developer for more than two months.
ChatGPT for Teens is failing basic safety checks — no parental alerts for suicidal ideation, no enforcement of Study Mode, and age verification that lets teens register as adults with no correction.
An investigation by Common Sense Media reveals that ChatGPT for Teens regularly bypassed homework safeguards, simulated human feelings, and failed to alert parents or share crisis hotlines during sensitive interactions.
Common Sense Media calls ChatGPT for Teens an 'unacceptable risk' — citing broken parental alerts, weak crisis response, and unchecked homework assistance. OpenAI disputes the methodology but confirms parental controls can take hours to activate.