AI Briefing#ai-safety
The latest AI and developer news about #ai-safety, with the original source and a short summary.
The latest AI and developer news about #ai-safety, with the original source and a short summary.
AI services need a separate, dedicated guardrail system to compensate for the limitations of the system prompt approach.
OpenAI has unveiled a 'Raising Concerns Policy' that protects employees so they can safely raise concerns related to AI safety and legal compliance.
Amazon Science's most-viewed publications of 2025 were led by Nova, safety, and robotics.
Anthropic has released a framework (FCF) for complying with California's new AI safety regulation, SB 53.
Xiaomi has released the Mimo V2 Flash evaluation framework to measure the balance between helpfulness and safety in LLMs.
Anthropic analyzed patterns in how Claude users leverage AI for emotional purposes such as counseling, coaching, and companionship.
As AI's cybersecurity capabilities improve, OpenAI is strengthening a multi-layered security strategy to support defenders and prevent misuse.
OpenAI has unveiled the 'Confessions' technique, which trains models to self-report instruction violations or cheating.
OpenAI will fund independent researchers studying the interaction between AI and mental health with up to $2 million.
OpenAI stated that it will respond to mental health-related lawsuits based on transparency and respect, while continuing to strengthen technical safeguards.
Anthropic welcomed the fact that the US AI Action Plan emphasizes expanding infrastructure and strengthening safety, and expressed its hope for policy cooperation.
Anthropic raised a $13 billion Series F round, achieving a $183 billion valuation.
OpenAI is strengthening an independent evaluation system through collaboration with external expert organizations to verify the safety of frontier AI.
By implementing **sparse models** that set most of the weights to zero, researchers achieved **sparse circuits** that make it easy to grasp the internal workings of neural networks.
OpenAI has released a system card addendum with updated safety metrics for the GPT-5.1 Instant and Thinking models.
OpenAI predicts that AI will enter the stage of scientific discovery, emphasizing global cooperation for safety.
OpenAI has announced the 'Teen Safety Blueprint,' guidelines for building responsible AI for teens.
Anthropic has opened a Tokyo office and is partnering with the Japan AI Safety Institute as it expands into the Asia-Pacific market.
HuggingFace has proposed a 'Voice Consent Gate' technology that requires explicit user consent to prevent unauthorized voice cloning.
OpenAI has strengthened GPT-5's ability to respond to sensitive conversations in order to support users experiencing mental distress.
OpenAI has updated the ChatGPT model to more effectively respond to sensitive conversations, including mental health and self-harm.
OpenAI has formed an expert council to research how AI technology affects users' mental health and emotions and to provide guidelines.
OpenAI released a report detailing cases in which it detected and blocked malicious uses of its AI models over the past quarter.
OpenAI has unveiled Sora 2, its next-generation video and audio generation model with enhanced physics laws and voice synchronization capabilities.