Criminals Hijack Cloud Accounts to Fuel AI Workloads, Inflating Bills

En bref

• Multiple 2026 cases involved hijacked cloud access used for AI workloads.
• Attackers drove API traffic and costs beyond $100,000 per day; one case reached $1 million.
• Google GTIG and Mandiant investigated; no CVE was involved.
• Watch for API surges, unexpected model usage, and billing anomalies.

During 2026, criminal actors gained entry to cloud tenants and leveraged API credentials to execute artificial-intelligence workloads, according to Google's GTIG and analysts at Mandiant. A report dated 3 November 2026 describes multiple cases in which intruders took over accounts and consumed AI services. No CVE was tied to the activity; the break-ins relied on valid access rather than a software flaw.

Victims are organizations with cloud accounts and model/API entitlements. In the documented cases, attackers generated unauthorized model usage and sharp increases in API calls, leaving inflated invoices. Daily costs exceeded $100,000 in some instances, and one case approached $1 million, the researchers said.

The pattern matters because AI compute is expensive and easy to trigger at scale. A compromised key can turn into a financial and operational incident within hours, and defenders may learn of it first through billing alerts. The absence of a CVE shifts attention to identity hygiene, key management, and anomaly detection.

Google and Mandiant's involvement indicates cloud providers and incident responders are tracking this as a criminal abuse pattern rather than a vulnerability campaign. Organizations should treat unusual model deployments, new API keys, and sudden request volume as possible signs of account takeover.

What to watch: spikes in API usage, unexpected AI service charges, changes to model access permissions, and credentials used from unfamiliar locations. Rapid containment of cloud identities and API keys remains the main control.

À faire maintenant

1. Audit cloud IAM roles, service accounts, and API keys now; revoke or rotate any credential that is unused, shared, or unexplained.
2. Configure budget and billing alerts for AI/API services, with thresholds tied to normal daily spend and escalation to security operations.
3. Monitor model invocation logs and API request rates; alert on sudden volume changes, new model deployments, or calls from unfamiliar regions.
4. Enforce MFA, least privilege, and short-lived credentials for cloud accounts; restrict who can create AI service keys or raise quotas.
5. Set hard quotas and rate limits on AI APIs, and require approval for high-cost model access.
6. Adopt a cloud-account-takeover playbook that covers key revocation, quota reduction, provider notification, and billing dispute steps.
7. Review recent activity for signs of unauthorized model use, including new keys, permission changes, and abnormal charges.

Source originale

Help Net Security