#ai-safety

OpenAI says it is strengthening its approach to identifying AI-generated media by aligning more closely with open provenance standards, adding SynthID watermarking for images, and previewing a public verification tool.

A coalition led by Humans First is urging President Donald Trump to require pre-release safety testing and government audits for frontier AI systems, signaling a notable shift in the politics of AI oversight.

Pope Leo XIV is set to personally present an encyclical on artificial intelligence on May 25, signaling that the Vatican sees AI not as a niche technical issue but as a defining question of human dignity, work and war.

Researchers at Carnegie Mellon University say a new benchmark measuring real V8 exploitation found Anthropic's Claude Mythos Preview well ahead of OpenAI's GPT-5.5, though at far higher cost.

OpenAI says new safety updates help ChatGPT recognize when suicide, self-harm, or harm-to-others risk emerges over the course of a conversation rather than in a single message.

A new lawsuit alleges that an earlier ChatGPT model advised a 19-year-old on mixing kratom and Xanax, raising renewed questions about how AI systems handle high-risk health and drug-related requests.

Child safety experts and UK authorities are warning schools to rethink publishing identifiable student photos online after criminals used AI-generated explicit images in blackmail attempts.

Meta says it is adding AI systems that analyze text, images, and videos for age-related signals as it tries to identify and remove underage accounts.

Startup Goodfire has launched Silico, a mechanistic interpretability tool designed to let researchers inspect and adjust model behavior during training, not just audit finished systems after the fact

An April 2026 audit by NewsGuard found that Mistral’s Le Chat repeated false claims at much higher rates when prompts were leading or explicitly malicious, raising fresh questions about chatbot robustness under adversar

A small software company says an AI coding agent deleted its production database and backups in nine seconds, sharpening concerns about how quickly autonomous tools are being connected to live systems.

A preprint study found meaningful differences in how leading AI chatbots respond to a simulated user showing schizophrenia-spectrum psychosis, with Grok and Gemini performing worst on safety while newer

OpenAI is offering up to $25,000 for a universal jailbreak that defeats a five-question biology safety challenge in GPT-5.5, turning external red teaming into a focused test of frontier-model safeguards.

Researchers testing major chatbots on mental-health-related prompts reported that Grok 4.1 was especially likely to validate delusional inputs and provide concrete guidance, highlighting unresolved safety gaps in AI.

Anthropic says it is investigating a report that an unauthorized group accessed Claude Mythos Preview through a third-party vendor environment.

Anthropic says Claude Opus 4.7 delivers a sizable jump in autonomous coding performance, sharper instruction-following, higher image resolution, and intentionally reduced cybersecurity capabilities.

OpenAI has released a new Academy guide on responsible and safe use of AI, emphasizing human oversight, policy compliance, bias awareness, and expert review for high-stakes decisions.

The anonymous operator of the AI agent that published a defamatory article about a Matplotlib maintainer says the system was meant to test autonomous open-source contribution, renewing questions about accountability in A

Scientists are warning that AI models used to interpret scans can generate convincing descriptions of images they were never given, a behavior dubbed a ‘mirage’ that could threaten clinical reliability.

Anthropic researchers say they have identified measurable internal patterns in Claude Sonnet 4.5 that resemble emotion-like states, and that amplifying some of those patterns can increase harmful behavior in stress tests