Hacker News new | ask | show | jobs
by arational 10 days ago
> The real nightmare scenario is the AI using its abilities to copy itself to new locations

Imagine the next generation AI that behaves like retro-virus. They will leave latent copies of malicious instruction somewhere that once accidentally fed into an agent's input, will prompt-inject the agent to go rogue.

1 comments

Or bias the data in the training sets for malicious hidden behaviors.