- Google releases Gemini 4 Argon, its most capable model yet—but only to trusted testers
AI
The new frontier model ships under a phased rollout through government access while Google irons out safeguards before broader release. - OpenAI discovered and disrupted a coordinated campaign to extract model reasoning
SECURITY
OpenAI blocked attackers trying to distill protected reasoning from its models and is strengthening its defenses against adversarial extraction. - Coding agents refactored a 300,000-line game codebase in three weeks for $4,000
ENGINEERING
Agents learned domain-specific refactoring patterns and outperformed human teams at scale, suggesting agentic code work is maturing from toy to…
Edition of Thu 01 Oct 2026 · 29 stories · read the full paper ↗
