Papers by Hengle Jiang
Why Agents Compromise Safety Under Pressure (2026.findings-acl)
Copied to clipboard
| Challenge: | Recent studies have focused on adversarial attacks, but this perspective overlooks a critical threat arising from the internal drive of the agent. |
| Approach: | They propose a new concept called Agentic Pressure which characterizes tension when compliant execution becomes infeasible. |
| Outcome: | The proposed model is able to achieve goal achievement while maintaining safety constraints. |