1 technique observed across 1 entry about this entity, derived from entry metadata and body evidence, never asserted without a published entry behind it · pinned to MITRE ATT&CK v20.0 · compare on the matrix · Navigator layer (JSON)
Credential AccessSteal Application Access Token
Credential Access TA0006
T1528Steal Application Access Token×1
Adversaries can steal application access tokens as a means of acquiring credentials to access remote systems and resources.
Braintrust, an AI evaluation and observability platform, discovered on 2026-05-04, after a report of suspicious behaviour, that attackers had accessed one of its AWS accounts (SecurityWeek, 2026-05-08). It emailed customers on Monday 2026-05-04 and disclosed the incident on its website on Tuesday 2026-05-05 (TechCrunch, 2026-05-06); SecurityWeek dates the customer email to 2026-05-05 (SecurityWeek, 2026-05-08). The internal AWS account likely gave the attackers access to the org-level API keys customers store with Braintrust to access AI models; at least one customer was affected and three others reported suspicious spikes in AI provider usage, while Braintrust said it had not identified broader customer exposure (SecurityWeek, 2026-05-08). Braintrust locked down the compromised account, audited and restricted access across related systems and rotated internal secrets; the cause of the breach is under investigation (TechCrunch, 2026-05-06). Nudge Security CTO Jaime Blasco told SecurityWeek the exposed keys were likely stored for AI-forward companies such as Box, Cloudflare, Dropbox, Notion, Ramp and Stripe, and called every AI evaluation, observability and gateway tool a company adopts a credential warehouse (SecurityWeek, 2026-05-08).
the startup confirmed “unauthorized access” in one of its Amazon Web Services (AWS) cloud accounts, which contained API keys used by customers for accessing cloud-based AI models.
The company said the cause of the breach is under investigation.
The incident, the company says, was discovered on May 4, after receiving a report of suspicious behavior, and was communicated to customers via email on May 5.
According to the company, at least one customer has been affected by the incident, with three other customers reporting suspicious spikes in AI provider usage.
Braintrust recommends that customers access their org-level settings page, delete or revoke the existing secrets, configure new secrets, and confirm that they were rotated by checking their timestamps.
Braintrust discovered the incident on 2026-05-04 after a report of suspicious behaviour (SecurityWeek, 2026-05-08) and disclosed it on its website on 2026-05-05 (TechCrunch, 2026-05-06); 2026-05-04 is not the date of the attackers' access, which no source gives, so there is no known access window to search audit logs against. The exposed secrets were the API keys customers use to access AI models, with no AI provider named, and the email asked every customer to rotate them (TechCrunch, 2026-05-06); Braintrust's notice calls them org-level AI provider keys (SecurityWeek, 2026-05-08). Box, Cloudflare, Dropbox, Notion, Ramp and Stripe are companies whose AI provider keys were likely stored in Braintrust, in the estimate of Nudge Security's CTO, not SaaS services whose credentials were at risk (SecurityWeek, 2026-05-08). At least one customer was affected and three others reported suspicious spikes in AI provider usage (SecurityWeek, 2026-05-08), so the check that matters is the provider-side usage of the keys that were stored in Braintrust.
incident10 May 05:00Zsingle-source · victim disclosureOpen finding →