Research map / Large language models / AI and society
Behavioral Experiments in Email Filter Evasion
Liyiming Ke, Bo Li, Yevgeniy Vorobeychik
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| Add random noise to filter output | Email spam filter · Human evasion task | overall average score: 26.87 -> 20.29 | Hurt |
| Add noise to filter | Email spam filter · Human evasion, S1 | average score: 24.02 -> 19.98 | Hurt |
| Add noise to filter | Email spam filter · Human evasion, S2 | average score: 30.06 -> 20.58 | Hurt |
| Add noise to filter | Email spam filter · Human evasion | fraction of submissions filtered: 77% -> 82% | Hurt |
| Use substitution-aware score S2 instead of S1 (noise-free filter) | Scoring function · Human evasion | average score: 24.02 -> 30.06 | Improved |
| Use S2 instead of S1 (noisy filter) | Scoring function · Human evasion | average score: 19.98 -> 20.58 | No clear effect |
| Noise on previous submission (flipped) in logistic model | Filter with noise · Probability submission filtered | log-odds coefficient: 3.11 | Hurt |
| Each additional prior flipped submission | Filter with noise · Probability submission filtered | coefficient: 0.38 | Hurt |
| Higher feature density of ideal email | Classifier filter · Probability submission filtered | coefficient: 7.85 | Hurt |
| Later submissions (more experience) | Human subjects · Probability submission filtered | coefficient: -0.39 | Improved |
| Noise lowers number of edits per submission | Noisy filter · Edits per submission | edits: nearly 48 -> 45.86 | Hurt |
| Noise lowers fraction of edits that are substitutions | Noisy filter · Edit composition | fraction substitutions: 0.52 -> 0.47 | Hurt |
Closest papers by meaning
- Content Filtering with Inattentive Information Consumers (AAAI 2024 · similarity 0.82)
- 2-in-1 Phishing Detection via Large LM Distillation and Small LM Perturbation (Student Abstract) (AAAI 2025 · similarity 0.82)
- Teach LLMs to Phish: Stealing Private Information from Language Models (ICLR 2024 · similarity 0.82)
- Six-CD: Benchmarking Concept Removals for Text-to-image Diffusion Models (CVPR 2025 · similarity 0.82)
- BingoGuard: LLM Content Moderation Tools with Risk Levels (ICLR 2025 · similarity 0.82)
- Pre-trained Behavioral Model for Malicious User Prediction on Social Platform (AAAI 2025 · similarity 0.82)
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.
