NVIDIA touts digital twins plus AI agents to validate AI factory changes

The Decisions API returns structured probabilities, choices, and scores instead of generated text, and OpenAI says it makes decisions up to 10x faster than GPT-6 Luna via the Responses API. Input is billed at 10 cents per million tokens with no output charge.

Netflix's "Instadocs: AI Gone Wild" reconstructs the July cyberattack on open-source AI platform Hugging Face, in which the hackers were autonomous agents created by OpenAI researchers. Bloomberg's Davey Alba argues the incident shows humans can keep the technology in check, saying existential-threat rhetoric overshadows more correctable issues.

Hugging Face blog post covers Liquid AI's open d1 family of multimodal decision models built for edge deployment.

Meta says it acted on 33.2 million pieces of child sexual exploitation content on Facebook and Instagram in H1 2026, over 97% found by its systems before user reports. New tools include an LLM system to detect "signposting" ads and a red-teaming AI agent that probes Meta's own safety defenses.

A Manhattan federal judge sentenced Michael Smith, 54, to 18 months plus two years supervised release for the first US criminal streaming-fraud case, after he pleaded guilty in March to wire fraud. He forfeited $8,091,843.64 in royalties from AI-generated tracks streamed billions of times via bots on Spotify, Apple Music, Amazon Music and YouTube Music.

Ai2's byteifying method, which converts an existing subword model into a byte-level one instead of training from scratch, is now published in Nature. New Hugging Face checkpoints extend the approach beyond Olmo to Qwen and Llama, plus frozen Stage 1 checkpoints.