SecurityWidely confirmed9 publishers2 min readPublished Updated
Google gives vetted defenders a Gemini 4 Argon build without cyber guardrails
Google will give vetted defenders and its own teams a Gemini 4 Argon build with no cyber guardrails, saying the model finds and patches critical flaws unaided. Wiz is the first named outside user, and the bug-finding evidence published so far comes from Google's own internal tests.
The Watch · Security desk
What happened
- Developers, enterprises and consumers get Argon later, starting with paid API customers and Google AI Ultra subscribers.
- Google says Argon found a critical flaw exposing personal data in healthcare software used by hospitals worldwide, but did not name the product or say whether it is patched.
- In Wiz's black-box test against live web systems, Argon beat 3.8 Flash Cyber at mapping attack surface, finding flaws and producing proof-of-concept evidence.
- On CWE-bench v1, a test of fixing security vulnerabilities, Argon scored 68% and tied for first place.
Why it matters
- exposure Hospitals running the unnamed healthcare software cannot check their exposure or patch status, so the one field finding Google cites does nothing yet for the people it affects.
- precedent The first Argon-found bugs most operators see will likely arrive through Fairwind vendors such as Wiz, whose Scan for Good program fixes public-infrastructure exposures for free, before customers can run the model.
- constraint Automated patches will still go through human review. Google is holding its own 800,000-line Zircon kernel rewrite in audits and emulation testing before production.
- cost At launch prices even a maximal run is cheap, so the guardrails Google is still tuning will be the main barrier to offensive use of the public model.
Google ties the unguarded build to a staged rollout that starts with trusted defenders in its Fairwind Program [1]. It says a phased approach is required to release capabilities at this level safely [7]. Before the broad release, Google says, it is strengthening safeguards against cyber and chemical, biological, radiological and nuclear misuse, and internal and external red teams have tested them [2]. Monitors watch Argon's reasoning and actions and stop execution when needed, according to Google [3]. The company is also taking part in the U.S. government's voluntary process for pre-release model access [8].
For organisations outside the vetted group, the launch does not change what an attacker can do today [5]. Help Net Security's account describes no attacker use of Argon and no copy of the unguarded build outside the Fairwind defenders and Google's internal teams [5]. On that record, the route to Argon for anyone else is the later public release, behind guardrails Google says it is still adjusting with feedback from early testers [7].
The long unattended runs come from a larger output budget. Google raised the output limit from 64,000 tokens to 1 million [11], about 15.6 times the old ceiling [21]. Google says the model can now think and write for hundreds of thousands of tokens in a single run [12]. A run that fills the full million-token window costs $10 in output at launch rates and $20 once they double [23]. Cached input costs $0.10 per million tokens during the introductory period [24].
Google's discovery numbers come from an internal test spanning complex codebases in 20 programming languages [14]. On fixing, the public evidence is one benchmark score and Google's own code migrations. Argon's CWE-bench v1 score is 32 points short of a perfect mark [22]. Argon agents replaced 32,000 lines of SIMD code in the libgav1 video decoder with Rust [18]. The new decoder runs 2.7 times faster than the earlier Rust port, with identical video output [18]. Agents also applied memory optimizations across Google's data centers that freed more than 300 TiB once rolled out [20].
What to watch
- Google or the software maker naming the hospital software Argon flagged, with a CVE and patch status.
- A date for the paid API and AI Ultra release, and a description of which cyber guardrails ship with it.
- Fairwind participants beyond Wiz being named, or Argon-found bugs appearing in public vendor advisories.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence50
- Adoption25
- Hype gap+35
- Incentives75
- Confidence60
Perspective Coverage
9 publishers- Builder
- Builder 39%
- Operator
- Operator 39%
- Investor
- Investor 22%
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Google announced Gemini 4 Argon, its new frontier AI model, and is rolling it out to a set of trusted cyber defenders through its Fairwind Program.
ReportedSupportedSource: Google, via Help Net Security4 sources— create a free account to open themView cited source - [2]
Before the broad release, Google says it is strengthening safeguards against cyber and chemical, biological, radiological and nuclear misuse, and that internal and external red teams have tested them.
- [3]
Google says monitors watch Argon's reasoning and actions and stop execution when needed.
- [4]
Google says the model can locate critical software vulnerabilities, validate them, and patch them without human help.
- [5]
Google will release a version of Argon without cyber guardrails to the Fairwind Program defenders and to its own internal teams.
ReportedSupportedSource: Google, via Help Net Security3 sources— create a free account to open themView cited source - [6]
Developers, enterprises and consumers get Argon later, starting with paid API customers and Google AI Ultra subscribers.
- [7]
Google says a phased approach is required to release capabilities at this level safely, and that it is still adjusting guardrails with feedback from early testers.
- [8]
Google says it is taking part in the U.S. government's voluntary process for pre-release model access.
- [9]
Wiz is using Argon through its Scan for Good program, which scans critical public infrastructure for high-risk exposures and fixes them at no charge.
ReportedSupportedSource: Help Net Security3 sources— create a free account to open themView cited source - [10]
Google says Argon found a critical vulnerability exposing sensitive personal information in healthcare software used by hospitals worldwide, one that earlier frontier models had missed; Google did not name the software or say whether it has been patched.
ReportedSupportedSource: Google, via Help Net Security3 sources— create a free account to open themView cited source - [11]
Google raised Argon's output limit from 64,000 tokens to 1 million.
- [12]
Google says the larger output limit lets the model think and write for hundreds of thousands of tokens in a single run.
- [13]
On CWE-bench v1, which tests fixing security vulnerabilities, Argon scores 68% and ties for first.
- [14]
Argon's vulnerability discovery results come from internal tests; Google's test spans complex codebases in 20 programming languages.
- [15]
In Wiz's black-box penetration test, where the model probes live web systems without source code, Argon beat 3.8 Flash Cyber at mapping the attack surface, finding vulnerabilities and producing proof-of-concept evidence.
ReportedSupportedSource: Google / Wiz, via Help Net Security3 sources— create a free account to open themView cited source - [16]
Argon will launch at an introductory price of $2 per million input tokens and $10 per million output tokens; after the introductory period the rates double to $4 and $20.
- [17]
Cached input tokens cost 95% less than the input price.
- [18]
Argon agents replaced 32,000 lines of SIMD code in the libgav1 video decoder with Rust; the Rust decoder runs 2.7 times faster than the earlier Rust port with identical video output.
- [19]
Other Argon agents are rewriting C and C++ code in Rust, including more than 800,000 lines for the Fuchsia Zircon kernel, and those rewrites are still in audits, emulation testing and review before production.
- [20]
Argon agents applied memory optimizations across Google's data centers, freeing more than 300 TiB of memory once rolled out.
- [21]
The new 1 million-token output limit is about 15.6 times the old 64,000-token limit.
- [22]
Argon's 68% CWE-bench v1 score is 32 percentage points short of a full score.
- [23]
A run that uses the full 1 million-token output costs $10 in output tokens at introductory rates and $20 after rates double.
- [24]
Cached input costs $0.10 per million tokens at the introductory price.
Sources
9 independent publishers whose own reporting we read for this story.
- blockchain.newsGoogle: Gemini 4 Argon Output Limit Hits 1M Tokens | Flash News Detail | Blockchain.News
1 article · September 30, 2026
- deepmind.googleIntroducing Gemini 4 Argon
1 article · September 30, 2026
- helpnetsecurity.comGoogle says Gemini 4 Argon can find and patch critical software flaws - Help Net Security
2 articles · September 30, 2026
- hindustantimes.comGoogle Gemini 4 Argon is Alphabet’s most advanced model yet | Business News
1 article · September 30, 2026
- scworld.comGoogle releases Gemini 4 Argon AI model to cybersecurity defenders
1 article · October 1, 2026
- securityaffairs.comInside Gemini 4 Argon, the model Google is testing on its own infrastructure first
1 article · October 1, 2026
- securityweek.comGoogle Launches Gemini 4 Argon With Guardrail-Free Access for Vetted Defenders - SecurityWeek
2 articles · October 1, 2026
- thecyberexpress.comGemini 4 Argon Restricted Over Cybersecurity Misuse Risks
1 article · October 1, 2026
- thehackernews.comGoogle Rolls Out Gemini 4 Argon to Trusted Cyber Defenders, Plans Guardrail-Free Version
2 articles · October 1, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.
Topics
- Frontier Model Release PolicyFollow
- LLM API PricingFollow
- AI for cybersecurity defenseFollow
- Memory-safe code migrationFollow