The past few days have offered a compressed preview of where AI's biggest fights are heading: politics, safety governance, and the economics of frontier models are all colliding at once.

Trump proposes an 'AI Force' — light on details, heavy on signaling

Over the weekend, President Trump announced plans to create an AI oversight body modeled on the Space Force, along with a yet-to-be-named "AI czar." In a lengthy Truth Social post, Trump said he would not let the “decimation” of AI happen, adding “we will not in any way hinder or stifle the Growth of this incredible Industry. Rather, we will cherish it, help it, and watch over it, as it grows!” He framed the initiative as a way to police bad actors using existing law rather than new regulation, and said on Truth Social, "For this purpose, I am forming the AI Force, much like I did Space Force."

The announcement landed just days after Nobel laureate Geoffrey Hinton warned lawmakers during a closed-door briefing that Congress has "maybe a year" to regulate the industry. Notably, Trump did not provide details about the new group or when he would appoint the adviser, leaving the substance of the plan almost entirely undefined even as it dominated AI-policy conversation this weekend.

Anthropic invites Accenture inside the building

A more concrete governance move came Friday, when Anthropic announced it would embed evaluators from Accenture's Faculty division directly inside its operations to stress-test model safety. According to the companies, Accenture's evaluators will red-team the AI lab's newest models and test their safeguards, with access levels inside Anthropic similar to the startup's own employees. The deal is substantial: the companies each expect to invest at least $1 billion over five years in AI safety.

The move operationalizes a pledge from CEO Dario Amodei's essay on slowing frontier AI deployment, and the partnership is not exclusive — Anthropic said it is also in discussions with the research nonprofit METR and other third parties. Still, the choice drew skepticism. More than 100 AI researchers, including Geoffrey Hinton, signed a public letter on the same day calling for embedded evaluators to be "meaningfully independent" — a pointed critique given that Accenture is already one of Anthropic's largest commercial partners, having agreed in December to train tens of thousands of its consultants on Claude.

StepFun's Step 5 undercuts frontier pricing by 7x

While Washington and San Francisco argued over oversight, Chinese lab StepFun quietly reset the price conversation. StepFun officially announced Step 5 Preview on September 20, a 600-billion-parameter sparse MoE with 27 billion active per token and a 1-million-token context, scoring 44 on Artificial Analysis's Intelligence Index — roughly a seventh the price of GPT-5.6 Sol. The model uses an unusual architecture: rather than widening the network, Step 5 Preview uses a 92-layer narrow-deep Transformer layout that the company argues gives longer information paths for implicit multi-hop reasoning during long prefill. Open weights are promised for mid-October, but for now the model is API-only.

Taken together, these three threads underline a familiar tension: as models get cheaper and more capable, the fight over who governs them — and how seriously — is only getting louder.

Sources

AI PolicyAI SafetyAnthropicArtificial IntelligenceDaily AI UpdateStepfun