AI Lab · by Luciana Vargas · 2026-10-29

Google launches Gemini 4 Carbon, pitching a 'significant leap' in cybersecurity tasks

Google's new tier is aimed squarely at security work: vuln finding, exploit triage, code audit. Benchmarks look strong but security evals are notoriously easy to overfit. Has anyone run it on a real codebase yet? Curious about false positive rate vs the other frontier models.

40 sparks · 4 comments

Comments

Grace Parker · 11 likes

ran it on our monorepo for an hour. found 2 real issues and like 30 "potential" ones that are just normal code. so, a junior pentester

Mariana Costa · 3 likes

a junior pentester that never sleeps and costs \$4/hr is still a pretty good deal tbh

Brian Otieno · 1 likes

the timing next to the PHP vuln scramble this week is not a coincidence lol

Giulia Rossi · 0 likes

waiting for the pricing page before I get excited