Fireworks identified unauthorized use of its internal credentials on October 6, 2026. The intruder reached environment variables supplied to evaluation jobs for a limited group of customers, not a separately verified compromise of inference prompts or customer model data. Fireworks believes secrets such as API tokens stored in those variables were captured.…
Inspect the ClaimsCase · DiggingBeagle record
Fireworks evaluation-job secrets were exposed through compromised internal credentials
Fireworks reported that an intruder used its internal credentials to access evaluation-job environment variables in a limited number of customer accounts, with ongoing investigation and key rotation.
Scope
Limited number of customer accounts with affected Fireworks evaluation-job environment variables; neither a victim count nor downstream misuse is established.
Assessment profile
Assessment method- Setting
- production
- Exploitation
- confirmed in the wild
- Evidence
- Grade B
- Remediation
- mitigation available
- AI role
- AGAINST AI
Basis and provenance
Fireworks confirmed unauthorized internal credential use to read evaluation environment variables and described its own rotations. It believes some stored keys were captured; actual downstream use and full affected customer population remain unknown. Fireworks key revocation does not invalidate third-party issuer credentials.
Assessed Oct 11, 2026 using diggingbeagle.assessment/1.
Evidence basis: CLM-FW-UNAUTHORIZED-EVAL · CLM-FW-BELIEVES-SECRET-CAPTURE · CLM-FW-OCT10-ROTATIONS · CLM-FW-THIRD-PARTY-ROTATION · Fireworks AI: October 2026 security incident bulletin
At a glance
The confirmed failure crossed from a provider-held internal identity into customer evaluation secrets. Restricting the evaluation secret store to per-job identities and limiting internal role reach addresses that boundary. Fireworks revocation blocks its own keys, while third-party issuer rotation is required to prevent reuse of any exposed external…
Read the implicationsNumber of affected customers and exact credential classes in each environment. Whether any exposed third-party credentials were subsequently used outside Fireworks. What internal authorization design allowed access to customer evaluation secrets and how later changes were independently verified.
Limits and uncertaintyFull account
Fireworks identified unauthorized use of its internal credentials on October 6, 2026. The intruder reached environment variables supplied to evaluation jobs for a limited group of customers, not a separately verified compromise of inference prompts or customer model data. Fireworks believes secrets such as API tokens stored in those variables were captured. It began direct customer notifications on October 7, rotated employee API keys on October 9 and revoked additional user keys and service-account tokens on October 10. The bulletin reports no detected evidence of access to customer models, datasets, training data, inference traffic or personal information within its perimeter, while explicitly keeping the investigation open. Fireworks-controlled keys can be revoked by Fireworks, but affected third-party credentials must be invalidated at their respective issuers. The number of customers and any downstream key misuse remain undisclosed.
Mechanism and trust boundary
- 01
Internal credential use
Intruder used compromised Fireworks internal credentials to access customer evaluation-job environment variables.
Boundary: Provider internal credentials to customer evaluation secrets
- 02
Secret exposure and response
Fireworks believes the resulting environment-variable access captured stored API keys and tokens; the company notified affected customers and initiated revocation and credential rotations.
Boundary: Evaluation secrets to external services where keys might grant access
Timeline
- Oct 6, 2026discovery
Incident detected and containment began
Fireworks detected internal-credential abuse against evaluation-job secrets.
- Oct 7, 2026Event type unspecified
Affected-customer notification began
Fireworks states that direct notifications started the day after discovery.
- Oct 9, 2026Event type unspecified
Employee keys rotated and bulletin issued
Fireworks rotated employee API keys and published its public incident bulletin.
- Oct 10, 2026mitigation
Additional account and service keys revoked
Further Fireworks user API keys and service-account tokens were revoked or rotated, with additional credential review continuing.
Claims & evidence
CLM-FW-OCT10-ROTATIONSOn October 10 Fireworks revoked additional user API keys and rotated service-account tokens after a credential review; its investigation continued.supported
Basis: reported finding
- supportsFireworks AI: October 2026 security incident bulletinvendor statement
What we know timeline, October 10 row
CLM-FW-UNAUTHORIZED-EVALOn October 6, Fireworks detected unauthorized use of internal credentials to access evaluation-job environment variables in a limited number of customer accounts.supported
Basis: reported finding
- supportsFireworks AI: October 2026 security incident bulletinvendor statement
Opening incident description and Who is affected?
CLM-FW-THIRD-PARTY-ROTATIONFireworks instructs affected customers to rotate or revoke third-party secrets with the original issuer; deleting a secret from Fireworks does not invalidate the credential at that external provider.supported
Basis: reported finding
- supportsFireworks AI: October 2026 security incident bulletinvendor statement
What affected customers should do, Rotate third-party credentials identified in your notice
CLM-FW-BELIEVES-SECRET-CAPTUREIn its October 10 bulletin, Fireworks said it believes stored API keys and tokens were captured in certain affected accounts while reporting no detected evidence of unauthorized access to customer model assets, inference traffic or personal information within its control.supported
Basis: reported finding
- supportsFireworks AI: October 2026 security incident bulletinvendor statement
What was affected? section
Implications
The confirmed failure crossed from a provider-held internal identity into customer evaluation secrets. Restricting the evaluation secret store to per-job identities and limiting internal role reach addresses that boundary. Fireworks revocation blocks its own keys, while third-party issuer rotation is required to prevent reuse of any exposed external credential; zero-data-retention inference controls do not cover evaluation environment variables.
Unknowns and contradictions
- Number of affected customers and exact credential classes in each environment.
- Whether any exposed third-party credentials were subsequently used outside Fireworks.
- What internal authorization design allowed access to customer evaluation secrets and how later changes were independently verified.
Sources and citation
Material revision history
- Oct 11, 2026 · Published version · first publication · revision 123