gandalf.lakera.ai
Learn prompt-injection concepts through interactive challenges. The current Maviter research brief focuses on the decision that matters most: Gandalf teaches prompt-injection behaviour through interactive challenges rather than serving as a production protection layer.
Start with a controlled decision, not a broad subscription. Complete several levels and document which instruction-hierarchy and data-exfiltration patterns each level demonstrates. Record attack-pattern recognition and ability to translate lessons into concrete controls. Review the official source, preserve the stated operating boundary, and keep the profile at review-pending status until Maviter completes independent evidence verification.
These checks follow the platform-specific brief; complete them before purchase or production use.
Start with a guided brief for Gandalf AI. Maviter keeps the selected platform in context and never exposes private member prompts here.
Commercial destinations are labelled and never influence editorial fit.
These official sources support a distinct working brief, but official documentation is not an independent benchmark. Pricing and consequential claims must still be checked before purchase or production use.
Adversarial testing for AI-generated software.
Automated code review, quality checks, and security analysis.
Automated AI review for every pull request.
Runtime protection for production AI applications.
Use the profile’s specific evaluation steps with representative, authorised material. Record what happened before deciding whether this tool fits your work.
Your notes stay in this browser tab. Refresh clears them. Nothing is submitted to Gandalf AI or saved to your Maviter account.