Skip to content

AutoDAN-Turbo

Automate jailbreak strategy discovery for red teams

Visit official site
Category AI models

What AutoDAN-Turbo is

The framework is described as an automated system that discovers and applies jailbreak strategies for LLM safety testing without human intervention.

What you can do with it

  • Automatic strategy discovery without human input
  • Compatibility with black-box and white-box LLMs
  • API compatibility with platforms such as OpenAI and Claude

Application scenarios

  • Test a newly developed LLM for effective jailbreak strategies
  • Integrate existing jailbreak strategies into a product test
  • Use it as a teaching tool for LLM safety evaluation

Pros & cons

✓ What we like

  • Existing human-designed strategies can be combined
  • An online learning mode explores strategies

! What to watch out for

  • The described setup names attacker, target, scorer and summariser LLMs
  • Runs are configured through a main script and parameters

FAQ

Does it need human-designed strategies?

The material says it discovers strategies automatically, and can also integrate existing human-designed ones.

What number is reported?

The material lists an average attack success rate increase of 74.3%.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review