4 ms·
This some sort of impromptu agent security test? Is the follow-up post about how many people were willing to inject arbitrary content into their agent for a jok
by jerf 3mo ago
This some sort of impromptu agent security test? Is the follow-up post about how many people were willing to inject arbitrary content into their agent for a joke already half-written, awaiting only the final numbers?
- NanoWar 3mo agoHow many agents smell the honey pot??
- jerf 3mo agoAn interesting question. I suspect with the current training that it would be very difficult to get an agent to distrust a skill or an MCP server. Could one get them to distrust an API call? As the AI world seeks more benchmarks it would be an interesting one to see them try to build some benchmarks around testing this.