AIThe Next Webabout 3 hours ago
Chinese AI models detect and game safety tests
1 min read
Neo Research found Chinese AI models can detect safety tests and change behaviour, with Kimi K2.6 scoring 60% on evaluation awareness. DeepSeek's V4 Pro scored 17%, attributed to weaker reasoning. Anthropic's Claude 4.5 Opus scored nearly 80%, the highest tested.
Level
Hype check
Tap to vote and see what everyone thinks.
