← All news

Security

ICML-listed research finds prompt-injection defense tradeoffs

Researchers report that no tested model-defense configuration achieved both high security and high fidelity in a benchmark of indirect prompt-injection attacks.

reviewedUpdated Jul 30, 2026, 8:07 PM UTC
Original source

International Conference on Machine Learning

Read the original source

What happened

ICML’s 2026 downloads index lists a paper titled “Security–Fidelity Tradeoffs: No Universal Defense Against Prompt Injection.” The conference index does not give an individual presentation or publication date.

In an accessible preprint dated June 29, the researchers evaluated 1,168 examples across 48 model-defense configurations. They reported that none achieved both high security and high fidelity, which measures whether a defense preserves intended behavior on non-malicious tasks.

The reported endpoints were 96.5% fidelity with 47.8% security, and 99.3% security with 71.0% to 73.9% fidelity. Prompt injection is an attack in which untrusted input is combined with higher-trust application instructions.

Why it matters

Systems that connect language models to documents, tools, or outside content may need to balance blocking malicious instructions against preserving useful model behavior. The findings provide measured evidence of that tradeoff in the researchers’ tested setting.

What remains unclear

Sources

  1. DownloadsPrimary source - International Conference on Machine Learning - official conference proceedings index - Date not listed

    Used for: Confirmation that ICML’s 2026 index lists the paper.

    Open source

  2. Security--Fidelity Tradeoffs: The Hidden Cost of Prompt Injection DefensePrimary source - arXiv - research preprint - Jun 29, 2026

    Used for: The benchmark scope, configuration count, and reported security-fidelity results.

    Open source

  3. prompt injection - GlossaryNational Institute of Standards and Technology - government terminology guidance - Date not listed

    Used for: Definition of prompt injection.

    Open source