Google restricts access to new Gemini 4 Carbon AI model for cybersecurity and coding

What's Happening

• Google DeepMind launched Gemini 4 Carbon, a frontier AI model designed for long-horizon agentic coding and cybersecurity defense.
• The model ranks first on the updated v5.0 Artificial Analysis Intelligence Index with a score of 59.
• Google restricts initial access to a vetted group of cyber defenders rather than offering a public API.
• Carbon outperforms rival models like GPT-6 Astra and Claude Opus 5.5 in 15 of 17 published benchmarks.
• The company also debuted Genie 4, a world model capable of simulating physical environments with audio.

Why It Matters

Google released Carbon two weeks after Argon, accelerating its frontier model cadence. The focus on autonomous engineering tasks shifts priority from general-purpose chatbots to specialized enterprise agents. Restricting access reflects safety concerns around AI systems capable of exploiting software vulnerabilities.