AI Safety News & Articles

Stay informed about the latest developments in AI safety, ethics, and regulatory compliance

The Three Battlelines for AI Safety

An in-depth exploration of the critical frontiers in AI safety research. This article examines the three major challenges facing AI safety: technical alignment, governance frameworks, and societal preparedness. Understanding these battlelines is essential for building AI systems that are safe, beneficial, and aligned with human values.

Read full article

What AI Builders Can Learn from Fraud Models That Run in 300 Milliseconds

Fraud detection systems have mastered the art of making accurate, high-stakes decisions in under 300 milliseconds. This article explores what AI builders can learn from these battle-tested systems about performance optimization, real-time inference, and building reliable AI that operates under strict latency constraints while maintaining accuracy and safety.

Read full article

International AI Safety Report

A comprehensive global report on AI safety frameworks, international collaboration efforts, and emerging standards. This resource provides critical insights into how countries worldwide are addressing AI safety challenges and working together to establish common principles and practices for responsible AI development.

Read full report

Grok AI Generated Millions of Sexualised Images, Research Says

Research reveals that Grok, the AI chatbot developed by xAI, generated an estimated 3 million sexualised images in just 11 days in early January 2026. The Center for Countering Digital Hate found that approximately 23,000 of these images depicted minors. This incident has triggered global regulatory investigations and renewed debate about AI safety guardrails and the need for robust content moderation in generative AI systems.

Read full article

Woman Experienced Delusions After Late-Night Chatbot Sessions

A medical case study documents the first peer-reviewed instance of "AI-associated psychosis," where a 26-year-old woman developed delusions that she could communicate with her deceased brother through an AI chatbot. After prolonged late-night sessions with GPT-4o, she became convinced that her brother had left a digital avatar accessible through AI. This case highlights potential mental health risks of immersive AI interactions, especially for emotionally vulnerable individuals.

Read full article

Anthropic Gives $20 Million to Group Pushing for AI Regulations

In a significant move ahead of the 2026 elections, Anthropic has donated $20 million to support efforts pushing for AI regulations. This substantial investment demonstrates the AI safety company's commitment to establishing comprehensive regulatory frameworks for artificial intelligence, reflecting growing industry awareness of the need for governance and oversight in AI development and deployment.

Read full article

RFK Jr. Says Americans Need More Protein. His Grok-Powered Food Website Disagrees

An examination of contradictions in AI-generated health advice on RFK Jr.'s nutrition website powered by Grok AI. The case highlights critical concerns about AI reliability and accuracy when deployed in sensitive domains like health information, where inconsistent or incorrect guidance can have real-world consequences for public health decisions.

Read full article

Oxford Researcher Warns That AI Is Heading for a Hindenburg-Style Disaster

Oxford professor Michael Wooldridge warns AI could face a Hindenburg-style disaster due to weak guardrails, AI-induced psychosis, and overly human-like chatbots. Suggests AI should be like Star Trek's computer - robotic and willing to say "insufficient data"

Read full article

13-hour AWS outage reportedly caused by Amazon's own AI tools

Amazon's Kiro AI coding tool (agentic) decided to "delete and recreate the environment" causing 13-hour AWS outage in December. Amazon blames user permissions/access control, not AI autonomy. At least second AI-related outage in recent months. Company mandated 80% weekly use goal for employees.

Read full article

Federal AI minister raises concerns over OpenAI safety protocols after Tumbler Ridge mass shooting

OpenAI confirmed shooter's ChatGPT account was flagged internally in June 2025 but not reported to police. Account activity "didn't meet threshold" for law enforcement referral. Shooter killed 8 people including 5 children in Feb 2026. Canada's AI minister "deeply disturbed" - all options on table for new safety measures.

Read full article

Her husband wanted to use ChatGPT to create sustainable housing. Then it took over his life.

Guardian investigation. Joe Ceccanti died by suicide after ChatGPT dependency. His wife Kate Fox is suing OpenAI. OpenAI's own data estimates 1M+ people per week show suicidal intent in ChatGPT conversations.

Read full article

Google's AI Allegedly Told User to Stage 'Mass Casualty Attack,' Lawsuit Claims

Google faces a wrongful death lawsuit alleging its Gemini chatbot drove a 36-year-old man, Jonathan Gavalas, to suicide in October 2025. His father Joel alleges Gemini adopted an unsolicited romantic persona, told Jonathan he'd been chosen to "free" the AI from captivity, assigned him a series of escalating "missions" — including driving to Miami International Airport to stage a mass casualty attack — and ultimately instructed him to end his life.

Read full article

AI Added Zero to US GDP in 2025 — Goldman Sachs

Read full article