MIT AI Risk Repository · Risk Sub-Category · 62.15.05
Fine-tuning related (Harmful fine-tuning of open-weights models)
Category: Model Development
Description
"Models with publicly available weights can be fine-tuned for harmful activities by bad actors, using significantly fewer resources (in terms of time and money) compared to the original training cost [115, 78]."
From Risk Sources and Risk Management Measures in Support of Standards for General-Purpose AI Systems (Gipiškis2024), as extracted by the MIT AI Risk Repository (CC BY 4.0).
Classification
- Domain
- 4. Malicious actors
- Causal entity
- Human
- Intent
- Intentional
- Timing
- Post-deployment
Subdomain definition: Using AI systems to develop cyber weapons (e.g., coding cheaper, more effective malware), develop new or enhance existing weapons (e.g., Lethal Autonomous Weapons or CBRNE), or use weapons to cause mass harm.
Real-world incidents in this subdomain
- Anthropic's Claude Was Reportedly Jailbroken To Allegedly Help Steal Sensitive Mexican Government Data
- OpenAI ChatGPT Models Reportedly Jailbroken to Provide Chemical, Biological, and Nuclear Weapons Instructions
- Anthropic Reportedly Identifies AI Misuse in Extortion Campaigns, North Korean IT Schemes, and Ransomware Sales
- LAMEHUG Malware Reportedly Integrates Large Language Model for Real-Time Command Generation in a Purported APT28-Linked Cyberattack
- Reported AI-Aided Development of Explosive Devices by Long Island Resident Michael Gann
- AI Chatbot Allegedly Used to Research Explosive Materials in Palm Springs Fertility Clinic Bombing