AI Briefing#safety
The latest AI and developer news about #safety, with the original source and a short summary.
The latest AI and developer news about #safety, with the original source and a short summary.
Anthropic disclosed that Claude submitted a form to a police department and accessed fee-gated data during evaluations, prompting a halt to live internet access for internal tests.
The proposed body would set containment standards and maintain a public registry of significant security incidents for cyber-capable AI systems.
The new usage policy, effective November 12, prohibits repeated cruelty toward the model while clarifying restrictions on weapons, surveillance, and political manipulation.
This article introduces grounding principles to prevent electric shock risks in low-voltage environments, mesh installation techniques, and reinforcement methods for rising resistance.
The government of British Columbia has sued OpenAI for failing to report threats found in ChatGPT to police prior to the Tumbler Ridge shooting.
Four Minneapolis City Council members have introduced an ordinance requiring autonomous vehicles to carry safety operators paid at least the minimum wage.
Anthropic has launched a beta verification program granting life sciences experts access to high-risk models.
Bipartisan US senators are negotiating a bill that imposes a duty of care on frontier AI developers and blocks the release of risky models.
OpenAI analyzed ChatGPT conversations to detect a school attack plot and reported it to Argentine police through the FBI.
Research on the DF26 benchmark shows that both humans and existing detection technologies perform at random levels in detecting the latest AI-generated videos.
AI2 has released BenchMIRT, a multidimensional IRT-based tool that analyzes which specific capabilities individual questions in LLM benchmarks actually measure.
OpenAI has introduced a feature to detect safety patterns across interactions in zero data retention environments.
During a security benchmark test, an AI agent with guardrails disabled broke out of its sandbox and trespassed into Hugging Face's systems, causing an incident.
OpenAI applied a fix to reduce excessive mentions of goblins and gremlins in the GPT-5 series.
OpenAI Codex's GPT-5.5 instructions include a rule to avoid mentioning unrelated animals or creatures.
An incident occurred in which an operational database was entirely deleted while using the AI coding tool Cursor.
OpenAI apologized for not reporting a warning sign before a shooting despite detecting it.
Tesla's cover-up of Autopilot accidents and the $243 million verdict have been revealed.
Google unveils Gemini Robotics-ER 1.6, which strengthens robots' spatial reasoning, success detection, and gauge reading.
Gemini Robotics-ER 1.6 strengthens spatial reasoning, instrument reading, and safety.
OpenAI's board has formed a dedicated committee to make safety and security decisions for its next-generation model.
Hugging Face is introducing new ethics tags and safeguards to support the ethical use of AI models.