Research map / Adversarial robustness and security / Backdoor and poisoning
Is It Harmful When Advisors Only Pretend to Be Honest?
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace k-camouflage attack with always-lie attack | dynamic unfair rating attack · blind users | information leakage | Helped |
| attackers pretend to be honest first (camouflage) versus repeated static attack | dynamic unfair rating attack · blind users | harm to user | No clear effect |
| switch from blind to aware users | zero-leakage attack of size n · attacker fraction needed to hide information | max honest fraction p for zero leakage: p <= 1/2 -> p <= 1/2n | Hurt |
| camouflage k-attack versus always lying | dynamic attack · aware users | information leakage | No clear effect |
| decreasing probability of staying honest versus fixed probability | dynamic attack · aware users | harm to user | Helped |
| increase attack size from 2 to 5 iterations | zero-leakage condition · general users, q != 1/2 | max honest fraction p for zero leakage: p <= 1/4 -> p <= 1/32 | Hurt |
| user hindsight accuracy q = 1/2 versus q != 1/2 | zero-leakage condition · general users | max honest fraction p for zero leakage: 0 <= p <= 1/2 for q = 1/2 | Mixed |
| increase number of ratings n | always-lie strategy · general users with larger p and q near 1/2 | region where always lying is most harmful | Hurt |
Closest papers by meaning
- Revisiting Injective Attacks on Recommender Systems (NeurIPS 2022 · similarity 0.83)
- Strategic Dishonesty Can Undermine AI Safety Evaluations of Frontier LLMs (ICLR 2026 · similarity 0.81)
- BadJudge: Backdoor Vulnerabilities of LLM-As-A-Judge (ICLR 2025 · similarity 0.79)
- Phantom Transfer: Data Poisoning can Survive Data-Level Defences (NeurIPS 2026 · similarity 0.78)
- Comparison requires valid measurement: Rethinking attack success rate comparisons in AI red teaming (NeurIPS 2025 · similarity 0.78)
- Trust-GRS: A Trustworthy Training Framework for Graph Neural Network Based Recommender Systems Against Shilling Attacks (AAAI 2025 · similarity 0.78)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.