AI evaluation and observability platform Braintrust has asked customers to rotate their API keys after hackers reportedly gained unauthorized access to one of the company’s Amazon Web Services (AWS) accounts. The incident raised concerns that sensitive AI provider credentials stored within Braintrust systems may have been exposed.
According to the company, the suspicious activity was detected on May 4 after receiving reports of unusual behavior. Braintrust informed customers about the breach on May 5 and shared indicators of compromise (IOCs) along with recommended remediation steps. The company stated that it immediately secured the affected AWS account, audited related systems, restricted access, rotated internal secrets, and launched a broader investigation into the incident.
The compromised AWS account reportedly contained organization-level API keys used by customers to access AI models and cloud-based AI services. Security researchers noted that the potentially exposed credentials may belong to companies using Braintrust for AI workflows, including organizations such as Box, Cloudflare, Dropbox, Notion, Ramp, Stripe, and others.
Braintrust told customers to revoke and replace all API keys stored within its platform “out of an abundance of caution.” While the company confirmed a security incident, it stated that there is currently no evidence suggesting widespread misuse of customer data or broader compromise beyond the exposed secrets. However, at least one customer was reportedly identified as impacted during the investigation.
Cybersecurity experts warned that compromised AI provider keys could allow attackers to misuse expensive AI services, access proprietary workflows, or generate unauthorized API usage charges. Some reports also suggested that several organizations observed unusual spikes in AI usage activity, potentially indicating that stolen credentials were already being exploited.
The breach adds to growing concerns around the security of AI infrastructure and SaaS platforms that manage third-party credentials. As enterprises increasingly integrate AI systems into their operations, attackers are increasingly targeting API keys, cloud infrastructure, and AI development environments to gain access to sensitive systems and services.
