Show HN: Locket – Robust feature-level access control for LLMs

Category: security

Tags: llm, access-control, adversarial-training

Score: 7.0/10 (Innovation: 8, Technical: 8, Documentation: 6, Utility: 6)

Locket introduces a novel feature-locking technique for LLMs, enabling pay-to-unlock schemes by training adapters that restrict access to specific model capabilities. The project is interesting because it combines adversarial training with model security for commercial LLM monetization.

Target audience: ML engineers, AI researchers

Repository: https://github.com/ssg-research/locket · Python · Apache-2.0 · 36 stars

View on Hacker News