AI policy moved from warning to enforcement this weekend as Minnesota's new nudification ban took effect over xAI's objection. At the same time, Google pushed foundation models deeper into robotics, OpenAI reset developer pricing, Snap tied recommendation to human creativity, and Nscale moved to own more of the AI cloud stack.
Minnesota's nudification ban survives xAI's last-minute emergency bid
Minnesota's ban on AI nudification tools took effect August 1 after a federal judge denied xAI's request for a temporary restraining order. The two-page order focused on timing: xAI filed on July 29, nearly three months after the law was signed and only three days before it was due to take effect. Judge Donovan Frank wrote that the delay cut against the claim of immediate harm.
That is narrower than a ruling that the law is constitutional. The court will treat xAI's request as a motion for a preliminary injunction, with briefing due through August 17 and a hearing scheduled for August 19. For now, however, the law creates an immediate test of how states can place responsibility on platforms whose tools can generate non-consensual synthetic nudity.
Gemini Robotics 2 connects reasoning to whole-body control
Google DeepMind has introduced three systems under the Gemini Robotics 2 banner. The core vision-language-action model controls full humanoids and bi-arm robots; ER 2 handles embodied reasoning, multi-step plans and robot teamwork; On-Device 2 runs locally and can adapt to new embodiments with a few hours of data.
The launch post shows robots walking, crouching, manipulating objects and collaborating. It also provides an important reality check. Google reports 92% success at unscrewing a bulb, but 36% at screwing one in, 44% at tying a rubbish bag and 32% using a dustpan. Those company-run results make this a substantial research release, not evidence of dependable general-purpose humanoids.
OpenAI moves the GPT-5.6 price floor
OpenAI has cut GPT-5.6 Luna API prices by 80% and Terra prices by 20%. From July 30, Luna costs $0.20 per million input tokens and $1.20 per million output tokens; Terra costs $2 and $12. Sol pricing is unchanged.
The same announcement replaces Priority Processing with Fast mode for Sol, which OpenAI says can run up to 2.5 times faster than Standard at twice the price. OpenAI says model-assisted kernel and serving work helped produce the savings. Those efficiency claims are its own, but the customer prices are concrete and materially widen the workflows that can run economically at high volume.
Snapchat makes human contribution a distribution signal
Snap says wholly AI-generated videos will no longer be eligible for recommendation on Spotlight. Its announcement frames the change as a response to repetitive, low-quality synthetic content and a way to prioritise authentic, human-made creativity.
The distinction matters. Snap is not banning every use of AI: videos enhanced or edited with Snapchat's own AI tools can remain eligible when transparency indicators are present. The company is using distribution rather than labels alone to shape creator incentives, turning meaningful human contribution into an algorithmic advantage.
Nscale buys further up the AI stack
Nscale has entered a definitive agreement to acquire Anyscale, the platform built by the creators of Ray. According to the announcement, the deal will bring roughly 200 Anyscale employees into Nscale and combine orchestration software with Nscale's data centres, power and GPU cloud.
Anyscale will keep its brand, and customers are expected to retain their choice of infrastructure. Financial terms were not disclosed, and the transaction remains subject to closing conditions and regulatory approvals. The strategic aim is clear: Nscale wants to compete as a vertically integrated AI cloud rather than a supplier of raw compute alone.
This edition spans policy, physical AI, model economics, creator distribution and infrastructure consolidation. The next signals are equally concrete: the August 19 xAI hearing, independent robotics tests, usage after OpenAI's price cuts, creator behaviour on Spotlight and whether Nscale can integrate Anyscale without weakening its cross-cloud appeal. Signal, not advice.