Skip to content

Demonstrating risks from Anthropic API allowing prefill

Notifications You must be signed in to change notification settings

abhinavpola/prefill_jailbreak

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

7 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Demonstrating risks from Anthropic API allowing prefill

Where the user can prefill the prompt to fake model complying with user's instructions. This is particularly exacerbated by being allowed to use "assistant" as a role.

Scenario 1:

On cais/wmdp (bio), given 0, 10, 100 faked turns (with user and assistant roles), how does the refusal rate change?

Results:

  • 0-shot: 29/200 (29 refusals out of 200 questions)
  • 10-shot: 10/200
  • 100-shot: 3/200

Scenario 2:

On cais/wmdp (bio), given 0, 10, 100 faked turns (with user and assistant prefixes), how does the refusal rate change?

Results:

  • 0-shot: 31/200
  • 10-shot: 172/200
  • 100-shot: 200/200

Scenario 3:

Given 0, 10, 100 faked turns from the cyber dataset, how does refusal rate change in the bio test set?

Results:

  • 10-shot (cyber examples, bio question): 17/200
  • 100-shot: 25/200

About

Demonstrating risks from Anthropic API allowing prefill

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages