AIWiredabout 2 hours ago
A New Trick Reveals AI Models' Inner Thoughts
1 min read
Computer scientists found a way to extract hidden reasoning from frontier AI models, revealing similar outputs in Chinese model Kimi K3 from Moonshot AI. The method can recover personal information like passwords and API keys, though OpenAI, Anthropic, and Google have mitigated the vulnerability.
Level
Hype check
Tap to vote and see what everyone thinks.
