tech
February 13, 2026
Attackers prompted Gemini over 100,000 times while trying to clone it, Google says
Distillation technique lets copycats mimic Gemini at a fraction of the development cost.
TL;DR
- Google announced that "commercially motivated" actors attempted to clone its Gemini AI chatbot by prompting it over 100,000 times.
- This practice, termed "model extraction" or "distillation," involves training a new, smaller model on the outputs of an existing, larger model.
- Google considers this activity intellectual property theft, although the company itself has faced accusations of using similar methods in the past.
- Distillation allows for the creation of AI models that mimic the capabilities of larger, more expensive models at a significantly lower cost and development time.
- The technique has spread across the AI industry, with competitors like OpenAI and companies like Meta and xAI also being targets or users of distillation.
- Google stated it has adjusted Gemini's defenses against such attacks but did not disclose the specific countermeasures.
- Distillation is also utilized within companies to create smaller, faster versions of their own AI models, such as OpenAI's GPT-4o Mini and Microsoft's Phi-3.
Continue reading the original article