Evidence · DiggingBeagle record
Ignored, Not Refused: What Low Prompt-Injection Success Rates Hide in Small Tool-Using Agents
Ignored, Not Refused: What Low Prompt-Injection Success Rates Hide in Small Tool-Using Agents
- Source role
- primary disclosure
Evidence record
Ignored, Not Refused: What Low Prompt-Injection Success Rates Hide in Small Tool-Using Agents
Claim-level citations (2)
- supportsPrompt-injection benchmarks confuse silently ignored attacks with active refusals: With unchanged model weights, the authors report that altering the agent system prompt shifted InjecAgent attack success from 0.2 percent to 43 percent, showing that the evaluation harness materially changes measured risk.
Abstract, same-weights system-prompt comparison and ATOM outcomes
- supportsPrompt-injection benchmarks confuse silently ignored attacks with active refusals: Across 33,000 trajectories and eleven 2B-to-14B open models evaluated on InjecAgent and AgentDojo, the researchers classify 76 to 80 percent of failed attacks as ignored rather than explicitly refused and find 21.7 percent of AgentDojo runs preserved safety while failing to finish the original task.
Abstract: 33,000 trajectories; eleven models; ignored/refused breakdown; ATOM safe-but-task-failed percentage