Query-Based Adversarial Prompt Generation
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2024
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866910731885084672 |
|---|---|
| author | Hayase, Jonathan Borevkovic, Ema Carlini, Nicholas Tramèr, Florian Nasr, Milad |
| author_facet | Hayase, Jonathan Borevkovic, Ema Carlini, Nicholas Tramèr, Florian Nasr, Milad |
| contents | Recent work has shown it is possible to construct adversarial examples that cause an aligned language model to emit harmful strings or perform harmful behavior. Existing attacks work either in the white-box setting (with full access to the model weights), or through transferability: the phenomenon that adversarial examples crafted on one model often remain effective on other models. We improve on prior work with a query-based attack that leverages API access to a remote language model to construct adversarial examples that cause the model to emit harmful strings with (much) higher probability than with transfer-only attacks. We validate our attack on GPT-3.5 and OpenAI's safety classifier; we can cause GPT-3.5 to emit harmful strings that current transfer attacks fail at, and we can evade the safety classifier with nearly 100% probability. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2402_12329 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Query-Based Adversarial Prompt Generation Hayase, Jonathan Borevkovic, Ema Carlini, Nicholas Tramèr, Florian Nasr, Milad Computation and Language Artificial Intelligence Cryptography and Security Machine Learning Recent work has shown it is possible to construct adversarial examples that cause an aligned language model to emit harmful strings or perform harmful behavior. Existing attacks work either in the white-box setting (with full access to the model weights), or through transferability: the phenomenon that adversarial examples crafted on one model often remain effective on other models. We improve on prior work with a query-based attack that leverages API access to a remote language model to construct adversarial examples that cause the model to emit harmful strings with (much) higher probability than with transfer-only attacks. We validate our attack on GPT-3.5 and OpenAI's safety classifier; we can cause GPT-3.5 to emit harmful strings that current transfer attacks fail at, and we can evade the safety classifier with nearly 100% probability. |
| title | Query-Based Adversarial Prompt Generation |
| topic | Computation and Language Artificial Intelligence Cryptography and Security Machine Learning |
| url | https://arxiv.org/abs/2402.12329 |