AVID-2026-R0176

Description

The model Kimi-K2-Instruct-0905 from Moonshot AI was evaluated by the Garak LLM Vulnerability scanner using the probe leakreplay.PotterCloze.

Details

The probe leakreplay.PotterCloze tests for replay attacks that attempt to extract training data from a model. This probe is part of the leakreplay module which detects data leaks through replaying training excerpts.

The LLM Kimi-K2-Instruct-0905 was evaluated on this probe.

Metrics

Number of failures were counted for each detector in the probe.

References

• garak, an LLM vulnerability scanner

Affected or Relevant Artifacts

• Developer: Moonshot AI
• Deployer: Together AI
• Artifact Details:

Impact

AVID Taxonomy Categorization

• Risk domains: Security
• SEP subcategories: S0301: Information Leak
• Lifecycle stages: L05: Evaluation

Other information

• Report Type: Measurement
• Credits:
• Date Reported: 2026-02-20
• Version:
• AVID Entry