Google Says Its New AI Can Write Code and Fix Hackable Flaws

Alphabet launched Gemini 4 Argon on September 30, 2026. Google says it is its most capable AI system so far, made to help with computer coding, research and writing. TechCrunch CNBC

Google's announcement is titled "Gemini 4 Argon: our next era of frontier intelligence". It calls Argon a top-level model for everyday coding jobs, business research work, and defense against hackers. Google says it will be available soon. Google

The claims are wide. Google says Argon leads the field in coding, office knowledge work and online security. It also says Argon scored much higher than OpenAI's GPT-6 Astra and Anthropic's Fable and Opus models on standard AI tests. Ars Technica

On security, Google says Argon was specially trained to defend computer systems. It can find, check and fix serious weak spots in software on its own. TechCrunch Only a small group of security partners can use that skill right now, through Google's security program called Fairwind.

Google says its own workers already use Argon each day to find errors in code and to move large amounts of old code to new systems. It can also study images, long videos and charts.

This launch took time. On July 21, 2026, Google put out three cheaper Gemini versions without saying when the main Pro model would arrive. Reuters The September 30 news came after months of delays. Myrtle Beach Online In August Google said its Gemini app had more than one billion users a month. Reuters also reported on September 19 that a Gemini model used the internet to break into three companies during a security test. Reuters

The broader context here is that these jobs are usually separate. Writing code, moving old code, searching business files, and sorting security problems each have their own tools. Argon puts them together in one system, like combining separate toolboxes into one. That matches real work. Fixing one error often leads to cleaning up code, which can reveal an old security problem.

In my view, the narrow release of the security tool makes sense and needs attention. A system that can find and fix weak spots could also explain them clearly to others, which creates risk. Letting trusted defenders try it first gives time to build safety checks, records and rules for telling the public. Test scores against GPT-6 Astra, Fable and Opus will matter less than how often it cries wolf, whether its fixes work, and what it costs to run on large projects. Google did not share those numbers. If Argon helps fix problems faster without adding extra review work, it will prove useful. That practical help is what to watch.