Business, technology and what comes next.
NextNow / Archive
OpenAI’s most capable model arrives with restricted cyber access, a phased enterprise rollout and new questions about how autonomous systems should be governed.
Nvidia’s $12.93 billion Hugging Face deal could reshape how businesses build AI. What its openness pledge means for costs, choice and enterprise control.
OpenAI, Anthropic, Google, Microsoft and security vendors issued a collective-defense framework aimed at companies, governments and frontier labs.
The disclosure offers creators and researchers a larger view of X’s recommendation system, though an important Grok-based rule-violation predictor remains closed.
New Claude models will add invisible marks to text and signed C2PA metadata to supported files as EU transparency rules take effect.
Sam Altman’s claim that AI has crossed a historic threshold is less prophecy than operating warning: Capability is compounding faster than institutions can absorb it.
Advanced models are gaining the ability to navigate systems, execute actions and circumvent safeguards. The institutions responsible for deciding how much risk is acceptable are still trying to determine what should be tested, by whom and what happens after a model leaves their control.
Autonomous AI agents are gaining permission to act across company systems. As that authority expands, liability is becoming a practical issue in contracts, cybersecurity, insurance and product design.
The first AI-generated viruses were engineered to kill bacteria, not people. That makes this experiment safer. It does not make the precedent benign.
OpenAI, Anthropic, Google and Meta were called to Washington to discuss a voluntary framework for government testing of the most capable U.S. AI models.
