Skip to main content
Back
IntrusionAzure
1 min read

Storm-2139 uses customer keys to generate images on Azure OpenAI

Victim
Microsoft Azure OpenAI customers
When
Activity documented by July 26, 2024Disclosed Jan 10, 2025
Entry
Exposed customer API keys.

Microsoft says a group it tracks as Storm-2139 used exposed customer API keys to generate harmful images on Azure OpenAI. The group also sold others access to the service. Microsoft revoked the access and pursued the operators in civil court.

What happened

Public sources
Exposed keys
Internet
Customer AI service
Azure OpenAI

Incident overview / Microsoft Azure OpenAI customers

  1. 1

    Customer keys become exposed

    Before discovery

    Microsoft says customer credentials were available from public sources. It did not establish how each key got there.

  2. 2

    Customer keys authorize access

    July 2024

    Microsoft’s complaint says the group used customer API keys to access Azure OpenAI accounts without authorization.

  3. 3

    Customer accounts generate images

    July–September 2024

    Microsoft alleges that the group used customer accounts to generate harmful images and made the service available to paying users.

Impact

Microsoft says the group generated harmful images using customer accounts and resold access to others.

Could this have been prevented?

Require identity-based access to Azure OpenAI

If every client can use Microsoft Entra ID, disable local authentication on the Azure OpenAI resource. The service will reject a request signed with a customer API key, even if that key is already public.

Incident control assessments

1
Prevented
0
Contained
2
Not covered

How it could have been prevented

The same replay with the controls in place. Green marks where a step is refused, blue where it is bounded, and red where no control reaches it. Open a row to read the control, why it acts, and its caveat.

Public sources
Exposed keys
Internet
Customer AI service
Azure OpenAI

Incident overview / Microsoft Azure OpenAI customers

1 · Customer keys become exposed

1 Not covered

2 · Customer keys authorize access

1 Prevented

3 · Customer accounts generate images

1 Not covered

Key facts

  • Microsoft discovered the misuse in late July 2024.
  • The affected keys belonged to Azure OpenAI customers.
  • Microsoft announced a court-authorized domain seizure in January 2025.
  • In February, Microsoft named four alleged developers in an amended complaint.

Sources

Every fact in this breakdown comes from the public reporting below. The replay is a model of that reporting, not a reconstruction of the environment it describes.

  1. 1.
    Microsoft Corporation v. Does 1–10, First Amended ComplaintMicrosoft, Eastern District of Virginia filing · Feb 27, 2025 · Vendor reportParagraphs 37, 79 and 140 identify the customer API keys, affected service, first observation and image generation.
  2. 2.
    Disrupting a global cybercrime network abusing generative AIMicrosoft · Feb 27, 2025 · Vendor reportAttributes the campaign, describes resale and identifies the named defendants.
  3. 3.
    Taking legal action to protect the public from abusive AI-generated contentMicrosoft · Jan 10, 2025 · Vendor reportDescribes exposure, revocation, civil action and the domain seizure.
  4. 4.
    Disable local authentication in Foundry ToolsMicrosoft Learn · undated · AdvisoryResource-level setting rejects authentication with API keys while permitting authorized Microsoft Entra identities.

The estate in the replay is illustrative and drawn at the level of attack stages; the victim's real environment is not public. Verdicts describe what a control does when it is in place before the incident. They are not an assessment of the victim's security program.

More cloud attack breakdowns

All breakdowns