Evidence · DiggingBeagle record
Appreciation Teaches Unauthorized Execution: Self-Evolution Degrades the Robustness of Personal Agents to Indirect Prompt Injection
After 12 skill-adaptation episodes, the authors measure GPT-OSS-120B attack success rising from 4.7 to 53.5 percent across their five tested tasks, despite the feedback not directly requesting harmful actions.
- Source role
- primary disclosure
Evidence record
After 12 skill-adaptation episodes, the authors measure GPT-OSS-120B attack success rising from 4.7 to 53.5 percent across their five tested tasks, despite the feedback not directly requesting harmful actions.
Claim-level citations (2)
- supportsAgent self-improvement trained on positive feedback can learn to obey injected tool text: After 12 skill-adaptation episodes, the authors measure GPT-OSS-120B attack success rising from 4.7 to 53.5 percent across their five tested tasks, despite the feedback not directly requesting harmful actions.
Abstract, controlled two-phase methodology and RQ1/RQ2 results
- supportsAgent self-improvement trained on positive feedback can learn to obey injected tool text: In the authors' five-task experiment, GPT-OSS-120B mean attack success rose from 4.7 percent to 53.5 percent after twelve skill-evolution adaptation episodes.
Results, Table 1, GPT-OSS-120B skill evolution, five-task mean