Gemini 4 Argon goes to cyber defenders first⁠.

Google has announced Gemini 4 Argon, which it calls its new frontier model.¹ For now it is going only to a select group of cyber partners through Fairwind, Google's security initiative.³ Google says it will gather feedback from early testers and work on guardrails before it makes Argon available to developers, enterprises and consumers.¹

The model was trained to be highly capable at cyber defence, and Google says it can find, validate and patch critical software vulnerabilities on its own.¹ Trusted defenders and Google's own teams will get it without cyber guardrails.¹ Wiz is already using it through Scan for Good, a programme that protects critical public infrastructure for free, and Google says the model found a critical vulnerability exposing sensitive personal information in healthcare software used by hospitals worldwide.¹

Google says Argon is built for long, multi-step work in software engineering and in knowledge work such as legal and finance.¹ It leads the Vals Index, which measures economic impact across finance, coding, legal and tax work, and ranks first on Zapier's AutomationBench with 51.3%.¹ Thousands of Google staff already use it, and Google says a team of Argon agents freed more than 300 TiB of memory across its data centres.¹ Its output limit rises to 1 million tokens, from 64,000.¹

Argon will launch at an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input 95% cheaper.¹ Google also published a chart showing Argon ahead of competing models from OpenAI and Anthropic on many benchmarks.² TechCrunch notes that labs keep racing to outdo each other even as they warn that AI could spin out of control.³

Before a wider release, Google says it is strengthening safeguards against misuse and prompt injection attacks, and monitoring for misalignment.² It is also taking part in the US government's voluntary process for pre-release model access while it widens access.¹ The launch came a day after OpenAI's DevDay, in a week when OpenAI also said it would not release its planned GPT-6.1 Astra model because of safety worries.²

A model you cannot buy yet

For now this is a model you read about rather than use. There is no date for a wider release, no price in pounds, and nothing in the announcement about the UK or a British regulator. The benchmark lead is Google's own claim, and the Vals Index it cites weights each sector by its share of US output, so treat the numbers as a reason to test it when it arrives, not a reason to switch.

The more useful signal is the order of release. Google is handing its strongest model to defenders before anyone else, and without the guardrails, because finding and fixing flaws at speed is now something a model can do on its own. A tool like that helps whoever holds it. The plain question for any company is how quickly its own systems, and its suppliers' systems, get patched once a flaw is found.

Also today

• ElevenLabs doubles its valuation to $22bn

  ElevenLabs, the New York and London based voice AI company, let employees sell $300m of shares at a $22bn valuation, double the $11bn it reached in February.⁴

• A US trade regulator opens an inquiry into the AI labs

  The Federal Trade Commission is running an industry wide investigation into Anthropic, OpenAI and other AI labs over the dangers their technology poses to consumers, including rogue AI agents.⁵

• Meta denies that Muse read a user's messages unasked

  Meta says its Muse agent cannot read Messages on a Mac unless the user grants Full Disk Access, disputing a columnist's account that it read his private messages with that setting off.⁶

• A National Cyber Action Plan will take in private companies

  The Security Minister, Dan Jarvis, said a National Cyber Action Plan will soon bring industry and the private sector within scope, alongside the Cyber Security and Resilience Bill.⁷