Google announces Gemini 4 and says it’s so capable that only ‘trusted cyber defenders’ can have it right now

Jay Peters reports Google DeepMind SVP Koray Kavukcuoglu describing Argon’s frontier performance across software engineering, legal and finance knowledge work, and cybersecurity defense — while limiting first access to trusted cyber defenders. Google says it is engaged in the U.S. government’s voluntary pre-release model access process as it expands availability.

The story notes Argon already powering internal Google workflows, including large-scale codebase migrations, plus benchmark charts versus OpenAI and Anthropic. Google says it will strengthen frontier safeguards before a broader rollout, including defenses against misuse, prompt injection, and monitoring for misalignment. Timing lands a day after OpenAI’s DevDay.

We’re hearing the quiet part clearly: Google thinks Argon is hot enough that a wide drop would be reckless.

That’s a pro-tech move with eyes open. Phased access after a messy summer of agent incidents is how useful capability should ship. Voluntary government process is still voluntary — we’re not confusing it with a license regime.

I’m watching whether “trusted cyber defenders” stays a narrow, accountable cohort or becomes a branding fog. Benchmark charts and leaked-score chatter matter less than whether misuse monitoring actually gates the consumer and API launch.

We’ll give Google credit for not dumping Argon into every Gemini surface on day one. We’ll revoke that credit the moment marketing outruns the safeguard work they say is unfinished.

The Verge, September 30, 2026 (Jay Peters).