EchoPrompt maintains cross-proxy robustness, averaging 87.06% AUROC across seven proxy models and outperforming DNA-DetectLLM by 2.70% AUROC on average¶

Claim · All claims
Evidence · 1 study · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? EchoPrompt remains strong across proxy model families and scales, though its performance is still influenced by the specific proxy family selected. → Hongrui Bao 2026

Evidence¶

Hongrui Bao 2026¶

Hongrui Bao, Yubing Ren, Jinhan You, Fang Fang, Shi Wang, Yanan Cao. (2026). ONCE A RESPONSE, ALWAYS A RESPONSE: DETECTING LLM-GENERATED TEXT VIA LATENT PROMPT RESTORATION. arXiv preprint. https://arxiv.org/abs/2608.05741 doi:10.48550/arxiv.2608.05741

q2 · i? · design · r2

Proxy-model analysis (Figure 5) across Qwen2.5-1.5B/3B, Llama-3.2-1B/3B, Falcon-7B, Llama-3.1-8B and Llama-3-8B. On Llama-family proxies EchoPrompt reaches 93.50% AUROC, exceeding IRM by 1.80% AUROC and 2.63% Best-F1.

"Figure 5 shows that EchoPrompt remains strong across different proxy model families and scales. Averaged over all proxy settings, EchoPrompt achieves 87.06% AUROC and 83.03% Best-F1, outperforming the strongest competing average baseline, DNA-DetectLLM, by 2.70% AUROC and 3.23% Best-F1."

Discussion¶

Related Claims¶

• EchoPrompt's AUROC rises with text length, from an average of 75.8% in the shortest 1-40 word bin to 98.9% in the 321-360 bin on DetectRL Length — related
• EchoPrompt remains robust across attack settings, obtaining the best AUROC and Best-F1 in four out of five attack groups and improving over IRM by 0.24% AUROC on average — related
• LLMs are more stable than supervised models across dataset shifts, though they do not surpass within-dataset fine-tuned performance — related
• The detection gain comes from restoring the missing prompt-response relation, not merely assigning an assistant role: the full prefix outperforms the empty prompt by 14.73% and 12.57% AUROC on Qwen2.5-3B and 5.63% and 5.33% on Llama-3.1-8B — related
• Detector accuracy declined significantly as text length increased, with both detectors performing best on short texts — reports the opposite
• LLM self-detection performance is strongly task-dependent: reliable for programming tasks and longer reflective responses but poor for short-answer questions — related