Defenders learned about compromises at three AI providers
Over five months, Dragos sent 2 million unique prompts and found security breaches at three providers. The study revealed widespread failures to protect customer data.
- Field
- computer security
- What they did
- Researchers sent nearly two million prompts with unique secret data to AI providers and monitored where those secrets appeared in the responses.
- Why it matters
- This revealed security breaches, prompt theft, and widespread failures to protect customer data.
#ai supply chain#canary tokens#data breach#dragos#prompt injection



