Silicon Angle reports that Google has begun rolling out its new frontier artificial intelligence model, Gemini 4 Argon, to vetted cybersecurity defenders. Currently available only to members of Google's Fairwind Program, the model aims to outperform competitors in various benchmarks.
Gemini 4 Argon, Google's latest AI model, is now accessible to select cybersecurity professionals through the Fairwind Program. Google claims Argon surpasses rival models from Anthropic and OpenAI on several benchmarks, including DeepSWE and AutomationBench, demonstrating strong performance in long-horizon software engineering tasks with an output limit of 1 million tokens. The model also exhibits improved resistance to indirect prompt injection and features separate monitors to prevent unintended actions. Google employees are already utilizing Argon for large-scale engineering projects, such as code migration and optimizing existing codebases, leading to significant performance improvements and memory savings, the company said. For instance, agents have rewritten critical code for a video decoder, resulting in a 2.7 times speed increase.
While a version with cyber guardrails removed is available to Fairwind members and internal teams, allowing it to identify and fix software vulnerabilities, Google is taking a phased approach to its release, emphasizing the hardening of safeguards against misuse. The company has not yet announced a release date for general developers and AI Ultra subscribers, but pricing details indicate it will be competitive with current market offerings.
Source: Silicon Angle
