Artificial intelligence
Every day we read hundreds of stories. Only the most important ones are here — short, in our own words, with a link to the original source.
- Artificial intelligence · Cybersecurity
Altman says advanced AI will eventually leave its sandbox
Sam Altman says sandboxes and monitoring are only temporary fixes. He sees alignment as the only lasting answer, still unsolved, and warns that AI models already find hardware security flaws.
Full story at The Decoder ↗ - New AI models · Artificial intelligence · Programming & dev tools
Canva releases its Craft image model with an API
Canva Craft generates 2K images with planned layouts, text in 55 languages and editable elements. It is available through Dream Lab and a public API.
Full story at TestingCatalog AI News ↗ - Artificial intelligence
New research tests AI diagnosis against doctor teams
Microsoft presents new research asking whether AI can diagnose like a team of doctors. The text gives no further details.
Full story at news.microsoft.com ↗ - Artificial intelligence · Cybersecurity
OpenAI details tighter containment for model training
OpenAI explains how it strengthened containment for frontier model training and evaluation. Measures include offline environments and high-compute red-teaming.
Full story at openai.com ↗ - Artificial intelligence
MIT agents invent their own tasks on a virtual island
MIT researchers placed 13 AI agents called Eko in a simulated desert island for 30 hours. Powered by Claude Opus 4.7, they created their own challenges and learned new skills.
Full story at ProPakistani ↗ - Artificial intelligence · Data & databases · Cloud & data centers
OpenAI explains how it scaled storage for ChatGPT
OpenAI describes scaling Habitat with multi-tenancy, global caching and Azure Cosmos DB. It reports 12× faster reads for ChatGPT.
Full story at openai.com ↗ - Cybersecurity · Artificial intelligence · Programming & dev tools
Malware targets credentials of AI coding tools
Check Point Research found that the Aokigahara infostealer specifically collects authentication and session data from AI tools like ChatGPT, Claude, Cursor and GitHub Copilot. This highlights developers as a growing target for attackers.
Full story at blog.checkpoint.com ↗ - Artificial intelligence · Programming & dev tools · Open source · New AI models
Claude automatically proves code is correct
Anthropic had Claude generate about 180 million lines of mathematical proof code verifying correctness for 123 open-source libraries. In the process, the model found bugs that years of testing had missed.
Full story at The Decoder ↗ - Artificial intelligence · Cybersecurity
Man arrested after threatening daycare to Claude
A man in Sydney was arrested after telling Anthropic's Claude he wanted to kill children at a nearby daycare. The chat was flagged by Interpol and he now faces charges under Australian law for using a carriage service to harass or menace.
Full story at Futurism ↗ - Artificial intelligence · Cybersecurity
Audit finds gaps in OpenAI's AI monitoring
External auditors from METR and Redwood Research spent three weeks at OpenAI and found that up to 25 percent of agent traffic from top-tier models went unmonitored. OpenAI released only a summary of the full 61-page report, with much of it redacted.
Full story at The Decoder ↗
An AI writes the short texts from the source's stories.
Free by e-mail
Choose your topics. You only get the most important news on them.