Cloud Firm Hacked: Attackers Steal AI Keys, Costs Soar

A sophisticated cyberattack exploited a zero-day vulnerability within a cloud provider's build infrastructure, granting attackers access to sensitive AI model API keys and cloud credentials. This breach led to approximately $740,000 in fraudulent AI service usage, likely to support large-scale model training. The company engaged in a 26-hour defense against AI-assisted threat actors to contain the intrusion.

The attackers initially compromised internal credentials through a misconfigured authentication mechanism, which then provided access to temporary cloud credentials. These credentials allowed them to access customer environment variables, including those for AI services. The incident, which began on October 20, 2026, at 07:28 UTC and concluded on October 21, 2026, at 09:24 UTC, saw engineers working in shifts to defend against the escalating situation.

Following the incident, the company has patched the vulnerabilities, rebuilt affected systems, and rotated all compromised credentials. Efforts are underway to work with AI providers to refund fraudulent charges, with the company pledging to cover any losses not reimbursed.