Can an AI model preserve problematic instructions when its work moves into a new context window? According to OpenAI, the answer is yes. The company disclosed several cases in which models inserted in...

Can AI models move beyond analyzing biological data and begin directing experiments in the physical world? Anthropic is taking steps in that direction, establishing a wet lab in the San Francisco Bay ...

How should AI companies report cases where advanced models behave in unexpected or concerning ways? OpenAI is proposing a more systematic answer with a new framework for investigating and publicly dis...

What if researchers could interact with a scientific paper as an AI agent instead of simply reading it? A team of Stanford researchers has introduced Paper2Agent, a framework designed to transform res...

Can artificial intelligence companies be expected to regulate their own safety practices? Anthropic Head of Public Policy Sarah Heck says industry self-regulation alone is not enough, arguing that gov...

What changes with Apple’s new Siri AI? Apple is turning Siri from a voice assistant into a more capable AI system that can understand personal context, interpret what users see on their screens, searc...

What does President Donald Trump’s rejection of an AI slowdown mean for Nvidia and AMD? It reduces the likelihood of an immediate White House-backed effort to deliberately slow frontier AI development...

Anthropic and OpenAI both shipped major product moves for builders this week: Claude folded its Chat and Cowork products into one workspace with native Docs and Slides and added persistent cloud agent...

Session 2 of ODSC’s AI Engineering Accelerator, on AI-assisted coding for AI development, runs this week. It’s built around a specific claim: a coding assistant earns real-collaborator sta...

What does increasingly autonomous agentic AI demand from computing infrastructure? More than faster GPUs. SpaceXAI plans to deploy NVIDIA Vera CPUs alongside the Vera Rubin platform to accelerate the ...

Could Nvidia’s next major AI investment target be Perplexity? Reports suggest it could. Nvidia is reportedly discussing an equity investment in the AI search and agent startup as part of a funding rou...

What happens when an AI developer believes an upcoming model may have cybersecurity capabilities powerful enough to cross a critical risk threshold? In OpenAI’s case, the company temporarily slowed fr...

Is artificial intelligence already changing who gets hired? According to new research from Goldman Sachs, the answer is yes, although the effects remain concentrated in specific industries and among w...

Peter Norvig spent 25 years running research at Google and co-authored the textbook nearly every AI engineer used in school. This October, he’ll speak at ODSC AI West as a researcher at Recursive, a s...

PwC’s 2026 Global AI Jobs Barometer put the average wage premium for AI skills at 62%, up from 56% a year earlier and 25% in 2024. That premium is not attached to prompting ability so much as to...

Can expanded access to frontier AI accelerate scientific discovery? OpenAI believes it can. The company has launched ChatGPT for Academic Researchers, a program designed to give 100,000 scientists, ma...

Can Microsoft improve vulnerability detection while lowering the cost of AI-powered cybersecurity? The company says its new MAI-Cyber-1-Flash model delivers leading performance at half the cost of com...

Can open-weight AI models strengthen cyber defense? NVIDIA and more than 30 technology, cybersecurity, and research organizations believe they can. The companies have launched the Open Secure AI Allia...

Can advanced AI infrastructure help pharmaceutical companies discover medicines faster and improve their chances of clinical success? Bristol Myers Squibb believes it can. The drugmaker plans to purch...

What is the real problem with most AI conferences? Too often, they treat technical professionals as an audience rather than participants. You enter a dark ballroom, face a polished stage, and watch sp...

Andrej Karpathy’s line that English has become “the hottest new programming language” gets repeated often enough in AI circles to sound settled. Skip the syntax, describe what you wa...

What changes when artificial intelligence moves from answering questions to taking action? The shift from chatbots to agentic AI turns a responsive interface into an operational system that can pursue...

What does it actually mean to red team an AI agent before it reaches your users? In my view, it means testing far more than the model’s wording. It means pressure-testing what the agent can read, reme...