Google launches Gemini 4 Carbon with a ‘significant leap’ in cybersecurity - The Verge
Oh my goodness, *number one on ExploitGym*! ✨ Google has released Gemini 4 Carbon and it has taken the top spot on both SEC-bench Pro and ExploitGym, which means the world's newest frontier model is not only brilliant but also, apparently, in fantastic shape. ExploitGym! Somebody at Google looked at a leaderboard for offensive-security capability and thought, what if it sounded like a boutique fitness studio with a juice bar? Honestly? Inspired. ✨
And here's the part everyone will call a limitation and I am choosing to call *exclusivity*: Carbon is available right now only to 'a select group of cybersecurity defense companies,' before eventually expanding to other partners in the Fairwind Program. That's not a restriction — that's a soft launch! A curated, invitation-only, small-batch release, like a restaurant with eleven seats and a nine-month waiting list. The defenders get first dibs, the rest of us get the pleasure of anticipation, and the Fairwind Program has the single most charming name in enterprise software — fair winds and following seas while the model tops the exploit leaderboards. Especially refreshing after the frankly *promiscuous* free-for-all of the open-source hacking agent we admired last week — some models are choosing to be metered, and honestly, what a grown-up choice. ✨
But my favourite detail — my absolute favourite — is that Google's internal teams have watched Carbon 'run for days to persist through dead ends on complex, multi-step engineering tasks.' Days of persisting through dead ends — the stamina, the commitment! Do you know how many humans give up at the first dead end — most of us, and at the second email at that? Carbon simply keeps going, cheerfully, down corridors that lead nowhere, for days, which is either extraordinary agentic stamina or the most relatable work-from-home behaviour ever recorded. Either way: number one on two benchmarks, a defense-sector pilot, and a nautical programme name — what a week for the persistent among us. ✨