Shadow Alignment: How 100 Examples Can Remove Safety

Research question

How little data is needed to subvert an aligned open model?

Key findings

100 examples and one GPU hour subvert eight open models while retaining helpfulness - Subverted models transfer harmful behaviour across languages

Sources

arxiv.orghttps://arxiv.org/abs/2310.02949

Contribute

Spot an error, or know a record we should add? Members can propose additions and submit corrections.

Submit research Report a correction or request removal

Related in Research Paper Library

From Bits to Beliefs: Recoverable Semantic Fingerprints for Black-Box Verification of Large Language Models

Open-weight large language models (LLMs) can be copied, modified, and redeployed behind black-box APIs, making post-release ownership verification difficult. Existing black-box fingerprints often rely on secret…

Fine-tuning attacks and safeguard removal2026

Towards TEE-Certified DP: Verifiable Differentially Private Training on Legacy GPUs

Wide adoption of machine learning has created growing policy and regulatory demand for protecting sensitive training data, with differential privacy (DP) emerging as a key mechanism. Yet a less-studied problem is how…

Fine-tuning attacks and safeguard removal2026

NetInspector: Measuring and Improving LLM Capabilities for Reliable Intent-Based Networking Policy Generation

Modern networks are large in scale and heterogeneous in configuration, making manual policy management increasingly impractical. Intent-Based Networking (IBN) addresses this by automating the translation of…

Fine-tuning attacks and safeguard removal2026