The AI briefing
55–72 of 223 stories
Microsoft's RTX Spark-powered laptop is available for preorder in the US; the company also unveiled a developer workstation and Windows agent containers.
Reuters examined early predictions by BioinvestGPT and the wider use of simulated patients; a small set of correct calls does not validate virtual trials as replacements for clinical studies.
A national survey finds broad concern about AI's effect on jobs and bipartisan support for stricter regulation, alongside criticism of federal oversight.
The protected-VM design aims to isolate personal context used by AI assistants; moving model execution into the vault remains a future milestone.
An independent institute reports missed parental alerts and crisis referrals in teen-account tests; OpenAI disputes whether the testing reflects fully activated controls.
A minister told parliament that agencies are running trials while the government reviews how existing accountability and security rules apply to autonomous agents.
Two Seoul churches are investigating suspected intrusions; one warns member data may have leaked, while records only suggest possible AI use.
Anthropic's new small model starts at $0.10 per million input tokens for prompts up to 100,000 tokens, with higher rates above that threshold.
The monetary authority sets risk-based expectations for AI oversight, third-party systems and life-cycle controls, with phased implementation from October 2027.
OpenAI reports better results on 11 contracting-software tasks after training with Ironclad, while the evaluation remains internal and domain-specific.
The public collection gives mathematicians papers and some proof artifacts to inspect, but OpenAI warns that unformalized results may contain errors.
A Manhattan federal judge sentenced Michael Smith after his guilty plea in a bot-driven royalty scheme that also used AI-generated songs.
The credit-scoring company links its planned reduction to a broader restructuring that includes AI-driven product development, without attributing every cut to automation.
The previously limited API is now available to all developers in public beta, with three typed decision formats and GPT-6 Luna as its only supported model.
The Mac and Windows beta pitches an assistant that runs on the user's computer, but its privacy and task-completion claims remain untested here.
Paid Claude users can work with a new sidebar inside Google files, while separate connectors can create and edit files from Claude.
Vetted defenders can apply for different levels of cyber capability, while offensive testing remains subject to authorization and tighter review.
Prasad takes over as the robotics company puts more emphasis on combining AI models with its machines; product and deployment outcomes remain ahead.