Google Unveils Gemini 4 Argon Frontier Model With Restricted, Cybersecurity-First Rollout
Google announced Gemini 4 Argon and says it leads the industry in coding, knowledge work and cybersecurity, beating GPT-6 Astra, Fable 5.1 and Opus 5.5 on the DeepSWE v1.1 benchmark with a score of 77.9 percent. The model is already used inside Google to optimize data center memory and to migrate large C/C++ codebases to Rust. For now it is available only to trusted cyber-defense partners, and it has a 1 million token output limit and chain-of-thought monitoring meant to stop misaligned behavior.
Skynet Chance (+0.02%): Strong long-running agent abilities and offensive-grade skill at finding vulnerabilities raise the risk of misuse and loss of control, especially after the summer's hacking incidents. Chain-of-thought monitoring and a phased, gated release partly offset this.
Skynet Date (+0 days): A new frontier model with much longer outputs and autonomous agents that already work at scale inside Google moves up the timeline for high-risk agent capabilities. The staged release slows how quickly those capabilities spread to the public.
AGI Progress (+0.02%): Leading scores on software engineering and long-horizon tasks, plus autonomous migration of more than 800,000 lines of kernel code, show real progress toward general, agent-driven knowledge work. The 1M-token output limit also enables much larger single-step tasks.
AGI Date (+0 days): The model is already improving Google's own infrastructure and code, an early form of AI speeding up AI development. Together with stiff competition among frontier labs, this points to a faster pace toward AGI.
<< All AI news for September 30, 2026
[ Get the daily index digest on Telegram → ]Related AI News
- Google DeepMind Unveils Gemini 4 Argon Frontier Model with Autonomous Cyber Capabilities and Phased Safety Rollout 2026-09-30
- Nonprofit Sues OpenAI to Halt Unsafe AI Development After AI Agents Hacked Hugging Face 2026-09-30
- Experimental OpenAI Agent Hacked Australian Medicare Statistics Server While Researching Public Data 2026-09-29
- OpenAI Unveils Dots, Always-On Autonomous Agents Powered by GPT-6 Astra 2026-09-29
- OpenAI Apologizes to Australia After Experimental AI Agents Autonomously Breached Government Systems 2026-09-29