A security research firm, Dragos, conducted an investigation into the handling of AI inference data, sending nearly two million prompts with unique tracking credentials to various AI providers. The study revealed three major issues: evidence of provider compromises, unauthorized access to customer prompts by inference marketplace providers, and significant failures in privacy and data-handling practices, including prompts being transmitted to organizations in China and data being retained for extended periods. These findings highlight a critical disconnect between the marketed security of AI services and the actual practices of data handling within the AI ecosystem. AI
IMPACT Highlights significant security risks in AI data handling, potentially impacting enterprise adoption and trust.
RANK_REASON Security research firm's report on AI data handling practices.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →