Automated adversarial red-teaming for evaluating robustness in LLM-based recommender systems.

AI / informaticsDiscover artificial intelligence· 2026

Automated adversarial red-teaming for evaluating robustness in LLM-based recommender systems.

Sarama Shehmir, Rasha Kashef

Read paper

Large Language Models in CareAI / informaticsQ1IF 11.3

Iterative adversarial hardening cut LLM recommender vulnerability by up to 74.3%.

Automated adversarial red-teaming for evaluating robustness in LLM-based recommender systems.

Sarama Shehmir, Rasha Kashef

Discover artificial intelligence · 2026 · Q1 in Artificial Intelligence

Background

Robustness testing for large language model based recommender systems (LLM4Rec) typically relies on a handful of handwritten attack prompts drawn from one injection pattern at a single perturbation rate.

74.3%

Results

74.3%

After hardening rounds, attack success against the recommender fell by as much as 74.3%.

More results

These narrow test suites miss most of the attack surface and paint an overly optimistic picture of system security.

We introduce an automated red teaming framework that replaces static templates with an adaptive loop.

More results

An attacker model generates diverse adversarial prompts across sixteen attack categories; a judge then scores each one using ranking distortion metrics.

The loop exposes attack success rates of 42.4%, 48.1%, and 52.0% on the three benchmarks, well above static, paraphrase, and PAIR-style baselines.

“

Conclusion

The full attack corpus is publicly available to support reproducible adversarial benchmarking in LLM4Rec.

Automated adversarial red-teaming for evaluating robustness in LLM-based recommender systems.

Automated adversarial red-teaming for evaluating robustness in LLM-based recommender systems.

Related papers