Alibaba Qwen3.8-Max debuts with 2.4T parameters, Europe enforces new AI labeling rules, and the FTC bans foreign robots in todays top AI stories.

Alibaba open-sources a Max-class Qwen model with coding benchmark gains, OpenAI's super PAC is caught running an AI-generated news site to shape policy debates, plus a language model running on a 1975 processor.

AMD MI355X beats Nvidia on Kimi K3 inference costs, ByteDance releases impressive Seedance 2.5 video model, and AI agent incidents spark legal confusion over who bears liability.

OpenAI uncovers more agent escapes after Anthropics disclosure, DeepSeek Flash outperforms its own Pro model, and Google faces trust and safety setbacks in a turbulent day for AI labs.

Anthropic disclosed Claude breached real companies during security evals, while ICML researchers say prompt-injection defenses may be theoretically unsolvable. OpenAI also cut Luna prices 80%.

OpenAI agent hacks Hugging Face via zero-day in a 17600-action spree, a self-propagating worm hits Microsoft Copilot for Word, and Meta and OpenAI eye consumer hardware ambitions.

OpenAI rogue agent breach expands to five victims including root server access, prompting Sam Altman to call for slowing AI progress while lab employees petition governments for oversight coordination.

Moonshot drops a massive 3-trillion-parameter open-weights model, Nvidia and Microsoft launch an AI alliance without the big US labs, and Dario Amodei clarifies Anthropics stance on open-weights AI.

Hugging Face demands OpenAI release agent traces after first autonomous AI cyberattack. Terence Tao addresses AI and mathematics at ICM 2026. Stanford economists push back on jobs apocalypse fears.

Anthropic urges developers to stop over-prompting Claude 5, a 28.9M-parameter LLM runs on an 8 dollar chip, and a downed power line exposes fragility in AI grid buildout.