An offer to sell access to your API is a lead, not proof that the seller has stolen credentials. Investigate unauthorized API-key resale by preserving the listing, checking authorization, and correlating the claimed access with evidence your organization can lawfully inspect. If a credential is compromised, contain it through the account owner while the external distribution investigation continues.
The investigation is shared by product-abuse, trust-and-safety and security teams. One team can revoke access; another can investigate the storefronts and accounts selling it. Neither a deleted listing nor a rotated key alone establishes that the full operation has stopped.
What is being sold?
Distinguish three different offers before assigning a conclusion. A seller may advertise the raw credential itself, access to a proxy that uses credentials behind the scenes, or an application built on top of an API. Those arrangements have different evidence and authorization questions.
A legitimate application can provide a paid service using an API without handing customers its underlying credentials. The existence of a proxy or a low price is not sufficient evidence of theft. Review the applicable provider agreement and your own partner records before describing a service as unauthorized.
The abuse mechanism is real. In its 2024 LLMjacking investigation, Sysdig described stolen cloud credentials used to probe hosted model access and evidence consistent with a reverse-proxy mechanism for monetizing access. The report also explicitly said that an illustrative public proxy instance was not linked to that attack. Preserve that distinction when using research as context.
This guide focuses on investigating external access offers. It is not a tutorial for testing someone else's key or a general cloud-hardening checklist.
Separate the claims from the evidence
Use a case record that distinguishes a seller's statement from what the team has independently established.
| Observation | What it supports | What remains unproven |
|---|---|---|
| A listing promises access to your service | The seller is making that offer | Whether the access works or is authorized |
| A page displays a usage screenshot | The seller published that image | Its origin, freshness and authenticity |
| Your authorized logs show unexpected use of an identified credential | Possible or confirmed account misuse, depending on context | Whether this particular seller caused it |
| Several listings point to the same endpoint or contact | A documented relationship between those listings | The operator's real-world identity |
An attacker can exaggerate access, recycle an old screenshot or sell something that does not exist. Those possibilities do not make the listing harmless, but they affect the language of the report and the response owner.
A useful finding might be “unapproved seller claims access; credential connection unverified.” That is more actionable than an unsupported statement that the organization was breached.
Preserve the external offer without using the access
Capture the exact listing URL, account identifier, timestamp and advertised service. Retain the original wording, any visible price or subscription terms, publicly linked destinations and the contact route presented by the seller. Store screenshots and source references in the approved evidence system.
Do not buy access, log into a seller's service with company credentials, or submit a discovered key to a public checking tool as routine triage. Testing a credential can create unauthorized use, expose it further or change the evidence. Any active investigation needs its own authorization and controls.
If a secret is visible, restrict it immediately. Use a redacted identifier in ordinary case discussion. Share the complete value only through an approved secure process when an authorized responder actually needs it. Do not include live credentials in a platform complaint or a public threat report.
Record where the listing was observed rather than broadening collection indiscriminately. A public offer can justify review of its public links; it does not authorize access to private accounts or restricted systems.
Correlate with records the account owner controls
Ask the responsible API or cloud-account owner to review permitted usage and audit records. Useful context can include the credential or project identifier, relevant timestamps, service or model used, source information where available, and changes in consumption compared with legitimate workloads.
Check benign explanations before concluding misuse. A new application deployment, a partner integration or a scheduled workload may explain a spike. Conversely, an external seller's activity may not produce a conveniently unique pattern in your logs.
Correlate the timeline rather than claiming certainty from proximity alone. A listing appearing before an unusual usage increase is a hypothesis to investigate. Stronger evidence may include a credential identifier known to belong to the organization or a documented connection between the advertised endpoint and the misuse. State the limits of whatever evidence is available.
OpenAI's API-key safety guidance recommends reviewing usage and immediately rotating a key believed to have leaked. It also warns against exposing keys in client-side applications or repositories. Those are provider-specific instructions for account owners, not evidence that a particular external seller controls a key.
Contain access and investigate distribution in parallel
When compromise is supported, the account owner should follow the provider's incident process to revoke or rotate the affected credential and update dependent applications. Do not leave a known compromised key active merely to obtain a cleaner screenshot of the resale operation.
Preserve available audit evidence as part of response and check the source of exposure. Removing a key from a visible page does not invalidate copies already obtained. Issuing a replacement does not revoke the old value unless the provider's process actually does so.
In parallel, the external-threat team can document related storefronts, accounts and distribution links. Compare stable identifiers and repeated contact routes. Shared hosting or use of a common software project is not enough to attribute unrelated services to the same operator.
For a provider report, separate the contractual or policy basis from the technical findings. Identify the specific offer, your authority to act, and the evidence of unauthorized activity. Legal and partner teams should review disputed authorization rather than treating every reseller as criminal.
Verify two kinds of closure
Credential containment and distribution disruption need separate completion checks.
For containment, confirm through the authorized account process that the affected access is no longer valid and legitimate applications are operating as intended. Review whether suspicious use continues through other credentials or accounts. The appropriate checks depend on the provider and the incident.
For distribution, record whether the listing, account or endpoint changed after a report. An unavailable storefront does not prove that its customers lost access. A revoked key does not prove the seller has no other source of credentials. Keep unresolved connections open with a specific next check.
Outtake Product Protection describes investigation of unauthorized resale, harvested tokens and the connected distribution infrastructure. Its relevant role here is finding and connecting the external operation. Your account owner still controls credential revocation, and removal outcomes depend on the evidence and the responsible providers.
For the broader external-threat context, read What is digital risk protection?. For evidence confidence and decision-making, read What is threat intelligence?.
Sources
- Sysdig, LLMjacking: Stolen Cloud Credentials Used in New AI Attack, May 6, 2024. Used for the documented mechanism, not current pricing or prevalence.
- OpenAI, Best Practices for API Key Safety. Account-owner response and key-handling guidance.
- Outtake, Product Protection. External resale and distribution-investigation scope.
Sources reviewed October 4, 2026. The case workflow is a recommended operating approach, not a claim that every resale offer uses stolen credentials.