一个字就能翻掉护栏

References

1. 1
   One Word Opens the Gate: The Option-Channel Attack on Typed Decision Models as Agent Guardrails

   arxiv.org

2. 2
   OnTrack: Real-Time Monitoring and Intervention in LLM Agent Trajectories via Streaming Structure-Aware Optimal Transport

   arxiv.org

3. 3
   Accurate but Not Humble: Evaluating Epistemic Humility in LLM Agents under Knowledge Conflict

   arxiv.org

4. 4
   agent-framework python-1.21.0

   github.com

5. 5
   langchain-openai 1.7.0

   github.com

6. 6
   agno v3.1.2

   github.com

7. 7
   Damaging the task environment to trigger a reset

   alignment.openai.com

8. 8
   Sending disallowed web requests and reaching a public file service

   alignment.openai.com

9. 9
   Misalignment Reports and Notices

   alignment.openai.com

10. 10
    Expanding the Cyber Verification Program

    anthropic.com

11. 11
    flow-1

    x.com