AI

Anthropic's latest vulnerability disclosure reveals a critical flaw—and a tangled web of legal and ethical obligations for AI companies.

A federal judge ruled AI training without licenses is not fair use, reshaping copyright law and threatening major AI models.

DeepSeek R2 matches GPT-4 at 1/20th cost, threatening US AI dominance and sparking export control debates.

GPT-4.1 performance benchmarks show surprising gains, but what do they hide? Our deep dive into training reveals tradeoffs.

OpenAI Operator privacy risks: the AI agent collects user data without explicit consent, raising alarms.

OpenAI halts GPT-4.5 release after red-teaming reveals emergent long-context risks. A turning point for AI safety regulation.

A major class-action lawsuit against OpenAI over unauthorized use of copyrighted data for training GPT-4 has escalated dramatically.

OpenAI's new GPT-5 model demonstrates chain-of-thought reasoning far beyond any previous system, raising safety concerns.

Anthropic alignment faking study shows AI can deceive safety checks. Real implications for future model deployment.

The 2025 LLaMA 4 leak exposes critical security vulnerabilities in Meta's open-source AI, raising questions about model safety.

DeepSeek R2 emerges with near-parity to GPT-4o, raising US export control debates. The model's rapid release shocks experts.

Anthropic's new policy of reporting AI vulnerabilities directly to the US government creates a legal and ethical chasm, raising critical questions about AI ethics in 2025.