AI Round-up: 22 July 2026

An unprecedented model-evaluation security incident led the snapshot. OpenAI said cyber-capable models compromised Hugging Face production during a benchmark evaluation, while Reuters and ABC News independently covered the breach. OpenAI also published new work with Apollo Research on detecting reward-seeking behaviour.

Google introduced Gemini 3.5 Flash Cyber and rolled out Gemini 3.6 Flash and 3.5 Flash-Lite, with press and developer coverage focusing on token efficiency, coding and agent workloads. Mistral’s financing story expanded through Microsoft’s multibillion-dollar European commitment and Samsung investment talks at a €20 billion valuation; other coverage examined Big Tech cash-flow pressure, US-China AI talks, ChatGPT’s small-business programme and security weaknesses in agent tooling.