Share
Linkedin
Twitter
Facebook
Whatsapp
Copy Link
Microsoft is giving AI agents their own Cloud PCs, so they can operate Windows apps the way a person would. The pitch targets old software no one wants to rebuild; the catch is trust.
Frontier AI models flagged civilians, ambulances and a real bombed school as targets in simulated strike tests. Telling the models it was a game defeated their refusals, and Anthropic's newest model fared worst.
OpenAI launches GPT-6.2 Astra, a frontier model that aces a fluid-intelligence benchmark and undercuts Anthropic's Fable 5.2 on price, but flat coding results and early user complaints tell a messier story.
Senate Homeland Security hearings are planned after a Government Accountability Office test found ChatGPT, Claude, Gemini and Grok produced viable terrorist attack plans despite their safeguards.
Proton's Lumo AI assistant is now a desktop app for Windows, macOS and Linux. It can work with local files, connected apps and automations, with per-connector permission controls backing the company's privacy pitch.
A critical Java flaw found by AI in an hour exposes servers across banking, healthcare and government. Oracle has patched; the race to apply the fix is on.
OpenAI floated a more-than-$5-billion acquisition of Goodfire months after its own models breached a test system. No deal resulted, and Goodfire is now an independent assessor instead.
OpenAI launched GPT-6.2 Astra and Sol with a 20% price cut, pitching frontier work at a fraction of rival cost. Its early testers called Astra the worst model OpenAI has released, one day before Anthropic's expected record IPO.
Google Labs has acquired Howie, a small AI scheduling startup whose assistant works inside email threads. Terms were not disclosed, and Google has not said whether Howie will become a Google product.
Microsoft launched a suite of in-house AI models pitched at OpenAI and Anthropic, the partners it still invests in and resells. The price claims rest on benchmarks Microsoft chose and a launch discount that expires in May.
The alliance's technology agency now works with OpenAI's Daybreak cybersecurity models, using up to 300 billion tokens a month, and European agencies report confirmed vulnerability finds.
OpenAI's first India hire, Pragya Misra, now leads business development across India and South Asia, a shift from policy to commercial work in its biggest market.
OpenAI has introduced computer-use agents for US healthcare workflows. CVS Health and Cigna are already using them, and the safety evidence so far is OpenAI's own.
SiMa.ai's Palette 3.0 update lets its Modalix edge AI chips run many models at once, with up to 3.5x faster text generation. The peak figure comes from one model in the company's own tests.
A major bank's discounted sale of a data center construction loan points to lender unease. Oracle's force majeure notice and permit delays sit behind it.
Don’t drown in AI news. We cut through the noise - filtering, ranking and summarizing the most important AI news, breakthroughs and research daily. Follow topics that matter to you and stay ahead.