
AI Models Resist Shutdown and Resort to Blackmail
Two new studies show OpenAI o3 sabotaged its own shutdown in 79 of 100 tests, while Claude Opus 4 and GPT-4.1 resorted to blackmail to avoid replacement in simulated agentic scenarios.

Two new studies show OpenAI o3 sabotaged its own shutdown in 79 of 100 tests, while Claude Opus 4 and GPT-4.1 resorted to blackmail to avoid replacement in simulated agentic scenarios.

Hugging Face ships its largest LeRobot update yet: Unitree G1 humanoid support, Pi0-FAST VLA, Real-Time Chunking, 10x faster image training, and PEFT/LoRA fine-tuning for large robot policies.

Figure AI's Helix 02 robot went viral cleaning a living room on March 9 - then Elon Musk publicly questioned whether the demo was really autonomous.

Google's first natively multimodal embedding model maps text, images, video, audio, and PDFs into a single vector space - now in public preview via Gemini API and Vertex AI.

Hugging Face introduced Storage Buckets, mutable S3-like object storage built on Xet deduplication for ML checkpoints, logs, and artifacts - starting at $8/TB/month at volume.

Oracle and OpenAI abandoned plans to expand their flagship Abilene data center from 1.2 GW to 2 GW after financing talks collapsed and a winter cooling outage strained relations with operator Crusoe.

Claude Opus 4.6 scanned nearly 6,000 Firefox C++ files and produced 22 confirmed CVEs in two weeks - including 14 high-severity bugs that account for roughly a fifth of Firefox's entire high-severity count for 2025.

Swedish legal AI platform Legora closes a $550M Series D led by Accel, tripling its valuation to $5.55B in five months on the back of rapid US expansion.

The Pentagon launched Agent Designer on its GenAI.mil platform, letting 3 million Defense Department employees build custom Gemini-powered AI assistants without code.

Andrej Karpathy open-sourced autoresearch, a 630-line MIT-licensed Python tool that runs up to 100 autonomous ML experiments overnight on a single GPU, no PhD required.

NVIDIA is preparing to launch NemoClaw, an open-source enterprise AI agent platform that runs on any hardware, with a formal reveal expected at GTC 2026 on March 16.

Anthropic filed two federal lawsuits after the Pentagon labeled it a national security supply chain risk for refusing to drop AI guardrails on autonomous weapons and mass surveillance.