Crowd-sourcing LLM Prompt Security
Prompt security isn't exclusive to enterprise; their users know them best. Craft secure AI prompts, challenge others by setting bounties on your vaults, and test your skills by attempting to breach vaults and claim rewards.
Craft, Test, and Harden Your AI's Defenses
Define the core instructions for your AI vault agent. Specify forbidden actions, optionally set a bounty, and test your prompt's resilience against adversarial dialogues designed to breach its security and claim the reward.
- Design detailed security prompts and rules for AI agents.
- Define specific forbidden actions (e.g., `allowAccess`).
- Set optional bounties on your vaults to challenge others.
- Attempt to breach other users' vaults to claim bounties.
- Evaluate prompt effectiveness based on security outcomes.
- Iterate on prompts to enhance AI security and robustness.
Frequently Asked Questions
Vault.ai is a platform for crowdsourcing prompt security. Users create AI-guarded vaults with specific security prompts and others attempt to breach them. This helps identify weaknesses and improves prompt security generally.