Key Takeaways

  • The attack turned a trusted third-party AWS integration into a path for cross-account access.
  • Exposed tokens and machine credentials can let attackers bypass conventional perimeter controls.
  • Short-lived credentials, narrow permissions, rapid rotation, and stronger monitoring can reduce exposure.

Trusted cloud connections are becoming part of the attack surface. In a recent supply-chain attack, attackers abused a third-party, cross-account trust relationship involving AWS, highlighting how an integration designed to simplify access can also provide a route into connected environments.

The central issue is not an AWS vulnerability. It is the trust granted to external identities, applications, and automation systems. AWS customers frequently authorize third parties to assume roles in their accounts for deployment, security monitoring, infrastructure management, and support. If an attacker compromises that third party or obtains its credentials, the attacker may inherit the access already approved by the customer.

That changes the defensive equation. A connection from a trusted role can look legitimate at the authentication layer, even when the activity behind it is hostile. Traditional controls focused on passwords, endpoints, or inbound network traffic may offer limited visibility because the attacker is using APIs and valid authorization paths.

How much access does an integration really need? In many environments, the answer changes over time. Permissions accumulate as teams add services, troubleshoot deployments, or expand automation. Temporary exceptions can become permanent policy. An integration initially intended to read configuration data might eventually gain access to secrets, storage, logs, or administrative functions across multiple accounts.

The wider secret-exposure problem makes these trust paths more consequential. Research reported by Cloud Native Now found more than 10,000 Docker Hub images exposing live API keys, cloud tokens, or CI/CD secrets. Of those images, 42% contained five or more secrets. The report also found that 75% of affected organizations that removed exposed material did not revoke or rotate the underlying credentials.

Public code creates another route. GitGuardian’s State of Secrets Report 2025 found that public GitHub repositories leaked 23.8 million secrets during 2024, a 25% year-over-year increase. Exposed AWS credentials have reportedly been probed by attackers in under 17 minutes. Deleting a token from a repository, container, or log is therefore only part of the response. Revocation and rotation matter too, and the window for acting can be very short.

Because cloud integrations are often machine-to-machine relationships, multifactor authentication (while valuable for workforce accounts) does not solve the risk posed by an overprivileged API key, federation assertion, or workload token. Defenders need to understand which non-human identities exist, what they can access, who owns them, and whether they are still active.

The CISA and NIST cloud-identity guidance released in 2026 reflects feedback from nearly 250 public comments. NIST IR 8587 identifies token theft, token forgery, secrets management, and API-based access as risks that extend across cloud providers. Its relevance is broader than federal systems because OAuth 2.0, OpenID Connect, and similar federation mechanisms underpin many enterprise integrations.

Detection also needs to follow credentials beyond a central secrets vault. HashiCorp has expanded Vault Radar to correlate values held in AWS Secrets Manager with exposures found in repositories, pipelines, logs, and container images. That approach addresses an awkward reality: a credential may be managed correctly at its source while copies appear elsewhere through build output, debugging, or developer workflows.

Practical controls include replacing long-lived keys with short-lived credentials, requiring external IDs for appropriate AWS cross-account roles, restricting role assumption conditions, and reviewing permissions against current operational needs. Organizations can also monitor AWS CloudTrail for unusual role assumptions, unfamiliar source locations, unexpected API sequences, and activity outside normal deployment windows.

Incident plans should treat a compromised integration as a potential multi-account event. Teams may need to disable role assumption, revoke sessions, rotate related secrets, examine downstream access, and coordinate with the third party. A trusted connection is useful precisely because it crosses boundaries. When attackers capture that trust, defenders need to investigate across those same boundaries, quickly and with enough context to distinguish routine automation from abuse.