Adobe's BPO Breach: One Phished Support Agent Extracted 13 Million Records and Unpublished Vulnerability Reports

A single customer support agent at a third-party business process outsourcing (BPO) provider was phished, gained elevated privileges, and used that access to extract 13 million Adobe customer records and unpublished vulnerability reports from Adobe's HackerOne bug bounty program.

The customer records (names, email addresses, account IDs, internal technical notes) are a significant data exposure. But the HackerOne submissions are the truly dangerous element. Unpublished vulnerability reports describe security flaws in Adobe products that have not yet been patched. In attacker hands, these reports become a roadmap for exploitation: step-by-step instructions for compromising Adobe software, written by security researchers and now available to criminal actors before patches ship.

The Third-Party Access Pattern

This breach follows the pattern that has defined 2026's most consequential incidents. The Vercel breach originated through Context.ai. The Mercor breach flowed from LiteLLM. The EU Commission breach was a customer-side cloud account failure. And now Adobe is breached through its BPO provider.

In each case, the organization that owns the sensitive data is not the organization that gets directly compromised. The breach occurs at a third party with legitimate access, and the data owner has limited visibility into the third party's security controls, employee vetting, or incident detection capabilities.

The Adobe breach adds a particularly alarming dimension: a single support agent was able to extract 13 million records in what appears to have been a single query or a series of queries that did not trigger volume-based alerts. When building the CIAM platform that served over a billion users, we implemented anomalous query volume detection on every data access pathway. A single account querying millions of records is not normal support activity. It is either a breach or a policy violation, and either way it should trigger an immediate alert.

What Makes This Breach Different

The theft of unpublished HackerOne vulnerability reports transforms this from a data breach into a security incident with cascading consequences. Each stolen report describes a vulnerability that Adobe's security team has not yet fixed. The attacker can use these reports to develop exploits for Adobe products before patches are available. The timeline advantage is significant: the attacker knows about the vulnerability, knows how to exploit it, and knows Adobe has not yet deployed a fix.

This creates a perverse incentive in the bug bounty ecosystem. Researchers submit vulnerabilities expecting them to be fixed. If a vendor's support infrastructure can be compromised to extract those submissions, the bug bounty program becomes a vulnerability feed for attackers. The trust model of coordinated disclosure depends on the receiving organization protecting the reports until a fix is deployed. Adobe's BPO breach broke that trust chain.

What Organizations Should Do

Implement data access volume controls for third-party agents. No single support agent should be able to query 13 million records. Rate limiting, anomalous volume detection, and per-session data export caps should be enforced at the application layer, not just the policy layer.

Segment vulnerability management data from support infrastructure. HackerOne submissions and security research should be accessible only to security team members through dedicated, separately authenticated systems. They should never be queryable through the same interfaces that customer support agents use.

Audit third-party access with the same rigor as internal access. BPO providers, managed service providers, and contractor organizations should be subject to identity governance reviews that match internal employee access reviews. The principle of least privilege applies to third parties as much as it applies to employees.

Enforce phishing-resistant authentication for all support system access. The breach started with a phished credential. FIDO2 security keys or passkeys for support platform access would have prevented the initial compromise.

Key Takeaways

• A phished third-party BPO agent extracted 13 million Adobe customer records and unpublished HackerOne vulnerability reports
• Stolen vulnerability reports provide attackers with exploit roadmaps for Adobe products before patches are available
• A single agent querying 13 million records without triggering alerts is an architectural gap, not a policy gap
• Third-party access is the dominant breach vector of 2026: Vercel (Context.ai), Mercor (LiteLLM), EU Commission (cloud), Adobe (BPO)
• Vulnerability management systems must be segmented from customer support infrastructure

Deepak Gupta is the co-founder and CEO of GrackerAI. He previously founded a CIAM platform that scaled to serve over 1B+ users globally. He writes about AI, cybersecurity, and digital identity at guptadeepak.com.