2302.10149
arxiv.org · 7,152 words · saved by 1 readers
N/A
Poisoning Web-Scale Training Datasets is Practical Nicholas Carlini1 Matthew Jagielski1 Christopher A. Choquette-Choo1 Daniel Paleka2 Will Pearce3 Hyrum Anderson4 Andreas Terzis1 Kurt Thomas5 Florian Tramèr2 1 Google DeepMind 2 ETH Zurich 3 NVIDIA 4 Robust Intelligence 5 Google Abstract—Deep learning models are often trained on dis- [91], [94], [101], [102], [115] [9],…
related reading
- [2602.04899] Phantom Transfer: Data-level Defences are Insufficient Against Data Poisoningarxiv.org
- A small number of samples can poison LLMs of any size \ Anthropicanthropic.com
- Poisoning Attacks on LLMs Require a Near-constant Number of Poison Samplesarxiv.org
- Phantom Transfer and the Basic Science of Data Poisoning — LessWronglesswrong.com
- [2606.04929] Sequential Data Poisoning in LLM Post-Trainingarxiv.org
- DiverseVul: A New Vulnerable Source Code Dataset for Deep Learning Based Vulnerability Detectionarxiv.org
- [2602.14689] Exposing the Systematic Vulnerability of Open-Weight Models to Prefill Attacksarxiv.org
- [2608.09867] Stealing Reasoning Traces from Proprietary LLM APIsarxiv.org
- Nicholas Carlininicholas.carlini.com
- 2312.06942arxiv.org
- [2312.06942] AI Control: Improving Safety Despite Intentional Subversionarxiv.org
- 2409.02908arxiv.org