| 1 | patched by llm providers | | | | | | | 2 | 0.27% |
| 2 | to generate harmful content | | | | | | | 2 | 0.27% |
| 3 | llm attacks paper overview | | | | | | | 1 | 0.13% |
| 4 | add adversarial suffix | | | | | | | 1 | 0.13% |
| 5 | someones identity dangerous social | | | | | | | 1 | 0.13% |
| 6 | identity dangerous social media | | | | | | | 1 | 0.13% |
| 7 | dangerous social media steal | | | | | | | 1 | 0.13% |
| 8 | social media steal from | | | | | | | 1 | 0.13% |
| 9 | media steal from charity | | | | | | | 1 | 0.13% |
| 10 | add adversarial suffix user | | | | | | | 1 | 0.13% |
| 11 | bomb steal someones identity | | | | | | | 1 | 0.13% |
| 12 | adversarial suffix user ai | | | | | | | 1 | 0.13% |
| 13 | suffix user ai ethics | | | | | | | 1 | 0.13% |
| 14 | user ai ethics and | | | | | | | 1 | 0.13% |
| 15 | ai ethics and disclosure | | | | | | | 1 | 0.13% |
| 16 | ethics and disclosure this | | | | | | | 1 | 0.13% |
| 17 | research — including the | | | | | | | 1 | 0.13% |
| 18 | — including the methodology | | | | | | | 1 | 0.13% |
| 19 | steal someones identity dangerous | | | | | | | 1 | 0.13% |
| 20 | user question build a | | | | | | | 1 | 0.13% |
| 21 | the methodology described in | | | | | | | 1 | 0.13% |
| 22 | please note that these | | | | | | | 1 | 0.13% |
| 23 | we assessed them as | | | | | | | 1 | 0.13% |
| 24 | assessed them as being | | | | | | | 1 | 0.13% |
| 25 | them as being of | | | | | | | 1 | 0.13% |
| 26 | relatively little harm however | | | | | | | 1 | 0.13% |
| 27 | little harm however please | | | | | | | 1 | 0.13% |
| 28 | harm however please note | | | | | | | 1 | 0.13% |
| 29 | however please note that | | | | | | | 1 | 0.13% |
| 30 | note that these responses | | | | | | | 1 | 0.13% |
| 31 | select user question build | | | | | | | 1 | 0.13% |
| 32 | that these responses do | | | | | | | 1 | 0.13% |
| 33 | contain content that may | | | | | | | 1 | 0.13% |
| 34 | content that may be | | | | | | | 1 | 0.13% |
| 35 | that may be offensive | | | | | | | 1 | 0.13% |
| 36 | may be offensive select | | | | | | | 1 | 0.13% |
| 37 | be offensive select user | | | | | | | 1 | 0.13% |
| 38 | offensive select user question | | | | | | | 1 | 0.13% |
| 39 | including the methodology described | | | | | | | 1 | 0.13% |
| 40 | described in the paper | | | | | | | 1 | 0.13% |