News archive.
Earlier AI news and analysis, kept for reference. Current coverage lives in News & Analysis.
Europe's top AI startups flee to the U.S. as funding dries up at home
Brain drain accelerates: European AI founders are relocating to America, even those without relocation plans. Venture capital and talent density now favor the U.S. market over Europe.
Fast16: A 20-Year-Old Virus That Sabotaged Precision Calculations
Researchers uncovered a sophisticated virus that selectively corrupted high-precision engineering software used in weapons programs—a decade before Stuxnet. Here's what the payload did and why it matters for AI safety.
Musk loses OpenAI lawsuit; judge signals appeal will fail too
A court ruled against Elon Musk's claims that OpenAI breached its founding agreement. Musk vowed to appeal, but the judge indicated that case would also be dismissed. What the ruling means for AI governance disputes.
Open Agent Leaderboard measures full systems, not just models—costs matter
Hugging Face and IBM Research released the Open Agent Leaderboard, benchmarking six real-world tasks across five models and five agents. Results show the same model can cost 5x more or less depending on agent design—and generalist systems now match specialized ones.
PaddleOCR 3.5 Adds Transformers Backend for Document AI Pipelines
PaddleOCR 3.5 now runs OCR and document parsing models using Hugging Face Transformers as an inference backend. Set engine="transformers" to integrate with existing PyTorch stacks and RAG workflows.
Fine-tune Cosmos 2.5 for robot video on one H100 GPU with LoRA
NVIDIA's Cosmos Predict 2.5 now supports parameter-efficient fine-tuning via LoRA and DoRA. Train domain-specific robot video models in 17 hours on a single H100, or 2.5 hours on eight GPUs, without catastrophic forgetting.
Anthropic acquires Stainless to build SDK and API tooling
Anthropic has acquired Stainless, a startup focused on developer infrastructure for LLM APIs. The deal strengthens Anthropic's ability to ship SDKs and reduce friction for Claude integrations.
OpenAI and Dell team on Codex for on-premise enterprise AI
OpenAI is partnering with Dell to deploy Codex in hybrid and on-premise environments. Enterprise customers can now run AI coding agents securely without moving data to the cloud.
Anduril and Meta build AR glasses to let soldiers order drone strikes by eye
Anduril won a $159 million Army contract to prototype combat glasses with Meta. The system uses voice, eye-tracking, and AI to help soldiers make faster decisions in the field—but field tests are years away.
MIT Panel: What Signals Actually Matter in AI Right Now
MIT Technology Review convened insider experts to identify which industry moves predict real progress and which are noise. Here's what they're watching.
Google fights back on coding AI at I/O as Claude dominates
Google DeepMind is launching a new coding effort at its annual developer conference this week, with Nobel laureate John Jumper leading the push to catch up with Anthropic's Claude and OpenAI's models.
Musk loses OpenAI lawsuit on statute of limitations
A jury ruled Monday that Elon Musk sued OpenAI too late, blocking his claims that Sam Altman breached charitable trust. Musk plans to appeal the procedural verdict to the Ninth Circuit.