Artificial Intelligence
Models, agents, and practical AI
OpenAI reorganizes its Model Spec around human intent, oversight and five values
OpenAI's October 29 Model Spec revision makes human intent and agency more explicit, while acknowledging that deployed models do not yet fully reflect the updated guidance.
Gemini 4 Carbon raises cybersecurity benchmarks while limiting early access
Google's new Gemini 4 Carbon reports stronger vulnerability discovery and patching results, but initial access is restricted to selected cybersecurity defenders.
Genie 4 makes AI-generated worlds interactive, but real-world fidelity remains unproven
Google DeepMind's Genie 4 adds live scene editing and generated audio to interactive world simulations, while short duration and visual drift limit current uses.
Anthropic’s public IPO filing puts AI growth and infrastructure costs under scrutiny
Anthropic has publicly filed for a proposed Nasdaq listing, offering a new view of rapid AI-company growth alongside substantial infrastructure expenses, accounting losses and unresolved technology risks.
OpenAI and Poland's NASK agree to explore joint AI safety research
A new agreement pairs OpenAI with Poland's NASK research institute to explore AI evaluation and safeguards, but measurable outcomes from the new collaboration remain ahead.