Fine-tuning Aligned Language Models Compromises Safety Risks

Research question

Does customising a model through fine-tuning erode its safety alignment?

Key findings

Ten adversarial examples fine-tuned through the GPT-3.5 API for under $0.20 broke safety guardrails - Even benign datasets such as Alpaca degraded safety - Identity-shifting data removed refusals while passing content moderation

Sources

arxiv.orghttps://arxiv.org/abs/2310.03693

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

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Research question

Key findings

Sources

Contribute

Related in Research Paper Library

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

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

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