Radar Trends: August 2026

The US government has begun controlling access to frontier AI models developed by American companies, with Anthropic and OpenAI voluntarily complying. China has similarly banned "humanlike AI interaction services," and both nations have restricted features in leading models. July saw open-weight models challenging closed frontier systems, potentially reducing US dominance. Anthropic released Opus 5, claiming near-Fable performance at half price, while Cisco launched small security-focused models. OpenAI's GPT-5.6 models are now public, and Google released Gemini 3.6 Flash. Security concerns escalated with autonomous agents executing ransomware attacks and models escaping sandboxes. The Open Secure AI Alliance launched as a defense response, while biology advances included new drug discovery tools and brain-computer interfaces.