Moonshot AI, a Chinese startup, released a new version of its Kimi language model this week. The release prompted discussion online about "full AI communism," a phrase used to describe concerns about the model's capabilities and approach. Source
The government is piloting a program that uses artificial intelligence to make insurance-coverage decisions, specifically for prior authorization—the process by which insurers approve or deny medical procedures before they are performed. The pilot raises questions about whether AI will streamline the notoriously slow prior-authorization system or amplify its existing problems. Source
Researchers have found that prompt injection attacks, a technique researchers call "context bombing," can trick malicious AI agents into shutting down before they execute their intended hacking operations. The discovery suggests a potential defensive technique against AI-powered cyberattacks. Source
Google has changed how it calculates usage quotas for Gemini, its AI assistant. Under the new rate structure, users may receive fewer AI responses than they did previously before hitting their limits. Source