Comet says attacker stole 29 of its own AI provider keys via Opik flaw

How the attack unfolded

According to Comet, the incident took place between October 24 and 26, 2026. An unauthorized party used a comet.com user API key from an inactive account to get into a small number of Comet's own internal test workspaces. Comet says this is the same kind of key any Opik user has and that it does not grant access to Comet's internal systems. The company states it has no evidence that its own systems were involved in how the key was obtained.

The account had no admin rights and no access to customer workspaces or data, Comet says. From the test workspaces, the attacker exploited a flaw in Opik's AI provider settings and obtained 29 API keys that Comet uses internally for OpenAI and Anthropic. The attacker also tried to use the same flaw to reach other Comet systems, but did not obtain any customer data, according to the company.

What was and was not affected

Comet lays out the scope in a table. Customer data in Opik, including traces, spans, datasets and prompts, was not accessed. No customer or user account was accessed or taken over. Customer AI provider keys were not exposed, since only keys stored in Comet's own Opik test workspaces were affected.

The one area marked as affected is Comet's own AI provider keys. The 29 keys for third-party AI providers were obtained and have since been revoked.

Response and fixes

Comet says an AI provider alerted it to unusual activity on October 26. The company contained the incident and blocked the exploited functionality the same morning, then deployed a fix. Its response included:

• Revoking all 29 exposed keys.
• Removing the inactive account from every workspace.
• Auditing all workspaces for stale members.
• Rotating all internal credentials as a precaution.
• Reviewing access logs for the affected workspaces and for every account the attacker created, to verify that no customer data was exposed.

While the fix was deployed, Opik on comet.com was unavailable for about two hours on October 26. Some features, including AI provider settings, the Playground and Ollie, were temporarily disabled.

Comet also changed how Opik handles provider settings. Opik now requires the API key to be re-entered when a provider address changes, and it restricts which addresses provider requests can reach. The company added monitoring for provider settings changes and unusual outbound traffic.

What users should do

Comet says users of Opik on comet.com, or of a dedicated deployment managed by Comet, need to take no action, because the fix is already in place.

Self-hosted and open-source Opik deployments were not affected by this incident, which was limited to comet.com. Comet says the underlying flaw did exist in previous open-source releases. Exploiting it requires an account in the deployment with permission to edit AI provider settings, and self-hosted Opik does not allow public sign-ups by default, so only invited people could do this. Even so, the company recommends upgrading to the latest version.

Customers with questions about how the incident relates to their account can contact their Comet account team or Comet support. Comet says it will update its post if anything changes.