VQV Signal
A preprint on arXiv measures eleven open-weights prompt injection guards on two corpora of legal contracts. The authors report that existing guards struggle, with document-level false positive rates as high as 92.39%. Two guards fall below chance on paired AUC. The paper notes that no false positive rate had been published for these guards on legal contracts, and the snippet cuts off during its comparison with a five-regex baseline.
A preprint on arXiv measures eleven open-weights prompt injection guards on two corpora of legal contracts. The authors report that existing guards struggle, with document-level false positive rates as high as 92.39%. Two guards fall below chance on paired AUC. The paper notes that no false positive rate had been published for these guards on legal contracts, and the snippet cuts off during its comparison with a five-regex baseline.
AI-assisted summary based on the listed source.
Specialized prompt injection guards are deployed in front of document analysis systems, but no false positive rate has been published for any of them on legal contracts. We measure eleven open-weights guards on two corpora of legal contracts and find that existing guards struggle: document-level false positive...
Teams that place these guards in front of contract review or document analysis systems may be relying on tools whose false positive rates on legal text were not previously published. The results suggest that guard performance on legal contracts should be tested directly before deployment. Watch for the full paper's baseline comparison and for any follow-up evaluations that test these findings.
Security-conscious readers may want to review the source and watch for practical exposure or mitigation details.
VQV organizes public signals from inspectable sources. It does not independently verify the underlying report.
Signal Strength reflects source quality, relevance, freshness and evidence. Public Interest helps organize discovery; it is not proof of truth.
VQV surfaced this signal because it is recent, relevant to AI Security, connected to arXiv.
No login, cookies, social SDKs, or automatic posting.
