An experiment by researchers revealed significant data leakage in AI services. By planting 2 million digital tripwires, they tested how sensitive information is managed and discovered vulnerabilities in the AI supply chain. The study highlights the risks associated with using AI brokers like OpenRouter and Vercel, which route user prompts to various inference engines. The researcher used unique credentials to identify potential leaks, underscoring the need for enhanced security measures in AI applications.
Key facts
- Researchers planted 2 million digital tripwires to test data leakage.
- The experiment revealed significant vulnerabilities in AI services.
- AI brokers like OpenRouter and Vercel were involved in the study.
- Researcher HD Moore used unique credentials to identify leaks.
- The findings emphasize the need for improved security in AI applications.
A short AI summary in our own words. The reporting belongs to Ars Technica; we don't reproduce it.