● HIGH↔ COMPANY DISPUTED
Researcher reports prompt-injection exploit caused customer-support assistant to disclose other users' PII
The named company has publicly disputed this characterization.
See the company response panel below.
- Date of incident
- August 4, 2025
- Jurisdiction
- US-WA
- Sources
- 2 cited
- Last verified
- March 29, 2026
Summary
A security researcher published a write-up alleging that a prompt-injection technique caused a SaaS customer-support assistant to surface portions of other users' personally identifiable information. The vendor disputes that any production PII was disclosed and states that the proof-of-concept used synthetic test data. The researcher subsequently agreed to coordinated disclosure terms.
What is reported
The researcher's August 4, 2025 write-up describes a prompt-injection technique in which a crafted user input caused the assistant to return text the researcher claims included PII.
Company response
The vendor disputes the characterization and states that the demo used synthetic data.
Sources
- 01
- 02