Google just built the most powerful AI model on the market — and you can't use it. Gemini 4 Argon, unveiled Wednesday, beats Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Astra on 12 of 18 benchmarks, but the public gets locked out while governments and corporations get first access. The same company that censored Americans at the Biden administration's request is now deciding who gets the tools that will shape every online conversation.

This is monopoly consolidation dressed up as innovation. Google isn't just selling a chatbot. It's building the infrastructure that will determine what Americans can see, say, and do online — and it's picking who gets in the door.

Argon goes first to Google's Fairwind Program, which launched in early September with more than 650 partner organizations, including CrowdStrike and Palo Alto Networks, along with governments and critical infrastructure operators. That version ships without cyber guardrails — the built-in refusals that normally stop a model from assisting with hacking, as Decrypt reported. The logic is that defenders need to think like attackers. The catch: the same skills cut both ways, and Google alone decides who counts as a "defender."

Google is also participating in the U.S. government's voluntary pre-release model access process, according to SiliconANGLE. Translation: Washington gets a look before you do. Koray Kavukcuoglu, Google's chief AI architect, wrote that releasing these capabilities "requires a phased approach." Phased for whom? Not for the 650 organizations already inside the velvet rope.

The benchmarks themselves warrant skepticism. Google computed its own DeepSWE v1.1 score of 77.9%, while rival numbers came from public leaderboards and company reports, as Decrypt noted. Argon leads on 12 of 18 benchmarks but trails on five. Bloomberg published a report the same day quoting a Google employee who said "the model struggles to handle certain coding tasks," according to Wccftech. On the Vals RSI index, Argon ranks fourth, behind three Anthropic models.

Still, the capabilities are real. Google says Argon agents already freed 300 TiB of storage across its data centers and helped Wiz find a critical flaw in healthcare software used by hospitals worldwide — one earlier frontier models missed. The model outputs up to 1 million tokens per response, up from 64,000. On indirect prompt injection attacks — where malicious instructions hidden in content hijack an AI assistant — Argon scored 0.7% compared to GPT-6 Astra's 8.5% and Grok 4.6's 51.8%.

When the public finally gets access, it'll cost $2 per million input tokens and $10 per million output tokens — introductory pricing that doubles to $4 and $20 later. Anthropic already charges those higher rates for Opus 5.5.

The pattern is clear. Google builds the model, gives the ungoverned version to government and corporate insiders, lets Washington vet it, and then offers the public a product with the guardrails bolted back on. Android Police framed the rollout as Google "strengthening safeguards" before public release. VentureBeat positioned it as Google "reasserting itself" in the AI race. Neither mentioned the obvious: the same company that spent years throttling search results and censoring content at federal request now controls the most capable AI model on earth — and it's letting the state touch it before any ordinary American can.

The question isn't whether Argon is impressive. It is. The question is who owns the models that will decide what you're allowed to see and say — and why the answer is never you.