Google and OpenAI complain about distillation attacks that clone their AI models on the cheap

2026-02-14 · Source: The Decoder · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy · Depth: Intermediate, quick

Summary

Google and OpenAI are reporting significant attempts to clone their large AI models, Gemini and ChatGPT, through a technique known as distillation. Google's Gemini experienced a single campaign involving over 100,000 requests aimed at extracting its internal logic, which Google characterizes as intellectual property theft. OpenAI has informed the US Congress about DeepSeek's alleged use of disguised methods to copy American AI models. Distillation involves flooding a target model with specific prompts to extract its reasoning steps, enabling the creation of cheaper clones that bypass substantial training costs. Google's John Hultquist warns that smaller companies are also vulnerable, especially if their models handle sensitive business data.

Key takeaway

For Directors of AI/ML concerned about intellectual property theft, recognize that distillation attacks are a present threat to proprietary models. Implement robust monitoring for unusual query patterns and high-volume requests against your AI systems. Your organization's competitive edge and data security depend on proactively defending against these sophisticated cloning attempts.

Key insights

AI model distillation poses a significant intellectual property theft risk, enabling cheap cloning by extracting reasoning steps.

Principles

• Distillation extracts internal model logic.
• Cloning bypasses billions in training costs.

Method

Distillation floods a target model with targeted prompts to extract its internal logic, specifically its "reasoning steps," to build a cheaper clone.

In practice

• Monitor for high-volume, targeted model requests.
• Protect models trained on sensitive business data.

Topics

• AI Model Distillation
• Intellectual Property Theft
• Large Language Models
• AI Security
• Model Cloning

Best for: VP of Engineering/Data, Director of AI/ML, Executive, AI Security Engineer, CTO, Policy Maker

Related on AIssential

• Anthropic accuses Alibaba of illicitly accessing its AI — AI model access controls are vulnerable to circumvention for competitive training via "distillation attacks".
• Detecting and preventing distillation attacks — Illicit AI model distillation poses significant national security risks and undermines export controls.
• Model Stealing Through the Lens of Model Multiplicity — High-fidelity stolen models may not be functionally equivalent to originals due to model multiplicity.
• Claude got attacked — Alibaba conducted a massive distillation attack on Claude, using 28 million exchanges to train its own model, prompting government intervention and potential KYC for AI.
• This "Dangerous" AI Model Just Got Hacked — Hyping an unreleased, powerful AI model as "too dangerous" can inadvertently increase unauthorized access attempts.
• AI Distillation Explained: The Truth Behind the Biggest AI Controversy Right Now — AI model distillation attacks are a significant and escalating threat to intellectual property and frontier model capabilities.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.