AIWiredabout 2 hours ago
AI Hacking Needs Human Guidance, Researcher Finds
1 min read
Researcher James Kettle found agentic AI is minimally capable but extremely limited at devising novel attack paths autonomously. Paired with human insight, AI proved powerful, uncovering a new vulnerability class called Shared-Parser Confusion. Experiments used Anthropic and OpenAI models starting September 2025.
Level
Hype check
Tap to vote and see what everyone thinks.
Summary by ByteBrief
