AI agents are harder to secure than we thought
Giving attackers more compute reliably uncovers more ways to hijack AI agents via injected instructions.
Giving attackers more compute reliably uncovers more ways to hijack AI agents via injected instructions.