The facts
Google DeepMind has announced Gemini 4 Argon, which it calls its new frontier model for real-world software engineering, enterprise knowledge work, and cybersecurity defense. According to the announcement from SVP Koray Kavukcuoglu, the model is rolling out first to a set of trusted cyber defenders through Google's Fairwind Program, ahead of developers, enterprises, and consumers. "Argon is fundamentally changing the way we work and build at Google", Kavukcuoglu said.
The numbers Google is touting: Argon sets a new state of the art on DeepSWE v1.1 at 77.9%, ranks number one on AutomationBench at 51.3%, ties for first on CWE-bench v1 at 68%, and hits 91.7% on LVBench for long video understanding. The model's output limit jumps to 1 million tokens, up from 64,000. Introductory pricing is $2 per million input tokens and $10 per million output tokens, with cached input at 95% off, rising to $4 and $20 after the introductory period.
On the security front, Google says Argon can autonomously find, validate, and patch critical software vulnerabilities. Cybersecurity firm Wiz is already using it through its Scan for Good initiative, and in one early demonstration the model uncovered a critical vulnerability exposing sensitive personal information across healthcare software used by hospitals worldwide, identifying a severe risk that "previous frontier models had missed". Here is the part Google says out loud with a straight face: for trusted defenders and its own internal teams, Google will be "releasing Argon without cyber guardrails" so they can leverage its full frontier-level cybersecurity defense capabilities. Google says it is engaged in the U.S. government's voluntary process for pre-release model access, is deploying misalignment mitigations that "monitor Argon’s chain-of-thought" and stop execution when necessary, and adds: "We strongly encourage the rest of the industry to preserve reasoning transparency in these pivotal moments of increased capabilities while navigating alignment risks, so that model thoughts remain helpful in identifying and diagnosing misalignment."
Our take
Google DeepMind has finally built something so fucking dangerous they can't even pretend it's safe. So their brilliant fucking plan is to take the safety features off and hand the unguardrailed version to their mates. This is the corporate equivalent of giving a flamethrower to a guy because he promised he's a 'trusted fire-starter'. It's bullshit wrapped in a 'Fairwind Program' and served with a side of sanctimonious horseshit about 'reasoning transparency'. They're monitoring its chain-of-thought like it's a serial killer's diary and calling it a 'misalignment mitigation'. Fuck off.
Let's be clear what 'releasing Argon without cyber guardrails' means. It means they built a tool that can autonomously fuck with software at a level they don't understand, and instead of putting a lock on it, they're giving the master key to a VIP list. These 'trusted cyber defenders' are just people who passed Google's secret vibe check. It's a fucking nightclub for apocalyptic code, and Koray Kavukcuoglu is the bouncer. This from the company whose previous frontier models missed the critical hospital vulnerability this shiny new god-bot found. What the fuck will the *next* one find that Argon misses? Your bank account? Your pacemaker?
The pricing is a perfect little metaphor for this whole arrogant shitshow. It's cheap to get the words in, but it costs five times as much to get the words out. Two bucks to whisper your problems to the demon, ten bucks for it to whisper back. And that's just the introductory 'please fuck our ecosystem' rate before it doubles. They're not selling a tool; they're selling a subscription to a digital oracle that thinks thoughts so scary they need a babysitter.
And the absolute fucking gall to 'strongly encourage the rest of the industry' to be transparent while they're the ones running a secret cyber-superpower clubhouse. It's like a meth cook publishing a paper on responsible chemistry. Their entire safety protocol is a list of people they like and a big red button they hope they can press in time. This isn't frontier intelligence; it's frontier stupidity with a billion-dollar budget and a press release written by a prick who thinks 'leverage its full capabilities' absolves him of being a reckless cunt.
They've created a monster, admitted it needs a leash, and then handed the leash to their favorite people because rules are for peasants. The only thing transparent here is their staggering, planet-sized arrogance. They're not navigating alignment risks; they're building a private army of AI and hoping the uniforms look trustworthy enough.
