/

CVE-2026-89049

Severity:

CRITICAL RISK

(9.9)

SSRF (CWE-918) in AWS SSM Agent Port Forwarding Function

CVE-2026-89049

Awaiting Analysis

Vulnerability type

Category:

Server-Side Request Forgery

CWE:

CWE-918, CWE-1289

Impact:

Cloud credential exposure and unauthorized access using instance role permissions

Severity:

CRITICAL RISK

(9.9)

CVSS version:

3.1

Catch this CVE while code is being written

Detect vulnerable code directly in pull requests, so CVEs are fixed before merge.

Summary

Amazon AWS Systems Manager Agent contains a server-side request forgery vulnerability in its port forwarding to remote hosts functionality. An authenticated remote user may bypass destination denylist checks using an equivalent representation of a blocked address, potentially reaching link-local endpoints and obtaining temporary IAM role credentials.

Affected context

The vulnerability occurs in AWS SSM Agent deployments that permit authenticated users to establish port forwarding connections to remote hosts. Risk is present when destination filtering relies on textual address comparisons and does not normalize equivalent representations of link-local or otherwise denied destinations before applying denylist rules.

Exploit preconditions

  • Attack surface: SSM Agent port forwarding to remote hosts
  • Required access: An authenticated remote user with permission to use the relevant port forwarding capability
  • Trigger point: Validation of the requested remote destination before a forwarding connection is established
  • Why exploitation is possible: Alternate address representations may evade denylist comparisons while resolving to a restricted link-local endpoint

Why this happens

The underlying bug is insufficient canonicalization and validation of network destination addresses. When security decisions compare an attacker-controlled host value without normalizing equivalent representations, a denied address may be expressed in another valid form and bypass filtering.

This maps primarily to CWE-918: Server-Side Request Forgery, because the agent can be induced to make a request to an attacker-selected internal destination. It also relates to CWE-1289: Improper Validation of Unsafe Equivalence, where distinct textual inputs are treated inconsistently even though they identify the same security-sensitive resource.

Potential impact

  • An attacker may reach link-local services that should be inaccessible through the forwarding feature
  • Temporary IAM role credentials may be exposed
  • The attacker could act with the managed instance's IAM permissions from outside the instance
  • Cloud resources and data may be accessed, modified, or deleted according to the compromised role
  • Further compromise or lateral movement may be possible in affected environments

General mitigation guidance

Mitigation generally focuses on canonicalizing destination values before authorization decisions and using a deny-by-default network policy.

Common defensive patterns include:

  • Resolve and normalize destination addresses before applying allowlists or denylists
  • Validate all equivalent address representations, including numeric, encoded, and mapped forms
  • Prefer explicit allowlists for permitted forwarding destinations
  • Restrict port forwarding permissions to users and roles that require them
  • Limit instance role permissions according to least-privilege principles
  • Reduce access to link-local metadata and management endpoints through network controls
  • Monitor and restrict exposure of temporary instance credentials

How teams detect issues like this

Detection can combine configuration review, authorization analysis, and network monitoring.

Useful approaches include:

  • Review destination validation for missing canonicalization before denylist or allowlist checks
  • Use static analysis and code review rules for comparisons against raw attacker-controlled host values
  • Test address-validation policies against equivalent representations and DNS resolution outcomes
  • Monitor SSM port forwarding activity involving link-local, metadata, or other internal address ranges
  • Alert on unusual requests for instance role credentials or cloud API activity from unexpected source locations
  • Enforce IAM and network policies that prevent forwarding sessions from reaching sensitive internal endpoints

Sources

SourceLink
aws.amazon.comaws.amazon.com/security/security-bulletins/2026-107-aws/
github.comgithub.com/aws/amazon-ssm-agent/releases/tag/3.3.4851.0
github.comgithub.com/aws/amazon-ssm-agent/security/advisories/GHSA-w9jw-h72g-6hxc

Frequently asked

questions

It may be exploitable remotely through the SSM port forwarding functionality when an attacker has valid authentication and the required permissions. It is not an unauthenticated network attack against the agent.

The attacker needs an authenticated identity authorized to use the affected remote-host port forwarding capability. The practical impact also depends on the instance role's permissions and surrounding network controls.

Normalize and validate destination addresses before making authorization decisions, use a deny-by-default or explicit allowlist policy, and prevent forwarding to sensitive link-local services.

Treat all destination values as untrusted, test equivalent address forms during security review, apply least-privilege IAM policies, and monitor forwarding sessions and access to cloud metadata endpoints.

Find out what's already
exploitable in your codebase.