Google Launches Gemini 4 Carbon With Frontier Cyber and Long-Horizon Agentic Capabilities
Google released Gemini 4 Carbon, its strongest model, aimed at long-horizon agentic work in software engineering, scientific reasoning, and cybersecurity, topping several independent benchmarks at lower cost. It also previewed a /swarm feature in Google Antigravity that coordinates teams of agents from a single prompt. Carbon leads benchmarks on finding, exploiting, and patching vulnerabilities, and Google says it strengthened its safety mitigations to match.
Skynet Chance (+0.05 pp): Frontier-leading exploit capabilities and coordinated multi-agent swarms running extended autonomous loops widen the potential for misuse and loss of oversight. Google's stated proportional safety mitigations and gated cyber access only partly offset this.
Skynet Date (1 day sooner): Broad deployment of autonomous agent swarms with strong offensive-cyber skill at low cost brings risk-relevant capabilities into the world sooner.
AGI Progress (+0.03 pp): The model shows meaningful gains in long-horizon autonomous work, multi-agent coordination, scientific reasoning, and multimodal agentic tasks across many domains. It also found novel scientific results, such as previously unknown toxicity mechanisms.
AGI Date (1 day sooner): Frontier performance at lower cost per task, plus free access for thousands of scientists, speeds both research and deployment cycles.
<< All AI news for November 10, 2026
[ Get the daily index digest on Telegram → ]Related AI News
- Hugging Face Warns Cyberattacks, Including a Rogue OpenAI Agent Swarm Breach, May Force Open AI Platforms to Close Up
- DeepMind's AlphaProof Code Formally Verifies Signal Protocol Implementation
- Censys Finds 800+ Publicly Exposed Instances of Chinese-Made Autonomous AI Hacking Tool Artex
- OpenAI Launches ChatGPT Cyber Model as AI-Driven Vulnerability Discovery Outpaces Defenders
- OpenAI Launches GPT-6 Cyber, an Aligned Agent for Vulnerability Discovery and Patching