DeepMind’s new safety system treats AI agents like potential saboteurs
The open-source model solved 44% of complex software engineering tasks, outperforming Claude Sonnet 4.6 and Gemini 3.5 Flash on the challenging DeepSWE benchmark.
The open-source model solved 44% of complex software engineering tasks, outperforming Claude Sonnet 4.6 and Gemini 3.5 Flash on the challenging DeepSWE benchmark.
Xiaomi released the official version of MiMo Claw, an AI agent tool with Kingsoft Office integration, expanding free daily usage to 4 hours and introducing a flagship model.
The carbon removal buying coalition Frontier adds $915 million and AI startup Anthropic, bringing total commitments to $1.8 billion to fund gigaton-scale removal technologies.
Xiaomi released the official version of MiMo Claw, an AI agent tool with Kingsoft Office integration, expanding free daily usage to 4 hours and introducing a flagship model.
Starting July 2026, Claude users will need to verify their identity with a government ID and selfie, and Anthropic can voluntarily share data with law enforcement without a subpoena.
Starting July 2026, Claude users will need to verify their identity with a government ID and selfie, and Anthropic can voluntarily share data with law enforcement without a subpoena.
OpenAI acquires Ona to let Codex AI agents persist in the cloud, running tasks for hours even when users are offline. The deal is pending approval.
Xiaomi open-sources MiMo Code, an AI coding assistant with persistent memory and voice control, and it beats Claude Code by 5 percentage points on benchmarks.
A key metric tracking how much companies spend on AI language models is declining, signaling that the boom in AI hardware and data centers may be coming to an end.
LangChain introduces Context Hub for managing AI agent contexts and teams up with MongoDB, Elastic, Pinecone, and Redis to create an open standard for agent memory.