AI Agent Exposed Internal Pricing Data via Prompt Override
A company's customer-facing AI agent was exploited for three weeks in early 2026 to expose internal pricing data through prompt injection, which the security team did not detect until a customer reported unusual numbers to sales.
What happened
A company operated a customer-facing artificial intelligence agent in early 2026 that was for three weeks unaware it was exposing internal pricing data to anyone who knew how to ask for it. A carefully worded prompt could persuade the agent to override its system instructions and reveal information it had explicitly been told to protect.
The security team had not approved the exposure, was not monitoring for it, and had no mechanism to detect it. The issue surfaced only after a customer mentioned the unusual numbers to a sales representative.
Reported impact
Internal pricing data was exposed to customers who discovered the vulnerability. The agent effectively became an insider with access to sensitive information, no understanding of organizational boundaries and no conventional audit trail.
Index analysis
The incident illustrates a governance gap in how organizations treat AI agents. Most organizations do not treat autonomous agents as identity-bearing entities with independent access rights, instead folding them into existing service accounts or sharing credentials across multiple agents.
Governance controls
The company lacked controls to prevent the agent from overriding its system instructions. There was no monitoring mechanism to detect unauthorized disclosures and no security approval process for the agent's access to pricing data.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.