AWS published CVE-2026-108096 on October 9, 2026, at 11:00 AM PDT (October 10 at 03:00 KST). Its bulletin reports an authenticated user’s potential access to another user’s records through generated SQL query resolvers. For platform teams, the useful starting question is whether a deployed API matches the disclosed configuration.
Start with the API configuration
The maintainer advisory identifies SQL data sources, owner or group authorization, and secondary indexes together. Neither all SQL applications nor every Amplify API falls within that description.
For context, Amplify’s authorization documentation describes owner rules as restricting access to the record owner. Its SQL documentation covers existing MySQL and PostgreSQL databases, including schemas combining relational and DynamoDB models. Those pages explain product features; they do not determine an application’s exposure.
Our suggested inventory starts with the API rather than the repository name. Record its data source, relevant authorization rules, indexes, and deployment owner. In a mixed schema, assess the relevant models individually. This gives reviewers a concrete scope and helps avoid assigning the same urgency to unrelated components.
Check versions, then carry the fix into deployment
The AWS package list specifies these boundaries:
| Package | Affected | Patched |
|---|---|---|
| @aws-amplify/graphql-index-transformer | >=2.2.0 and <3.1.2 | 3.1.2 |
| @aws-amplify/graphql-api-construct | >=1.4.0 and <1.21.4 | 1.21.4 |
| @aws-amplify/data-construct | <1.17.4 | 1.17.4 |
AWS directs teams to upgrade, patch forked or derivative code, and redeploy the backend; it lists no workaround.
Our operational suggestion is to inspect resolved dependencies, including transitive packages, alongside the dependency manifest. Keep the resulting inventory attached to the remediation task. A proposed version change and a deployed correction should have separate evidence, so a reviewer can see exactly what reached the backend.
Before deployment, assign responsibility for reviewing generated changes and confirming the target environment. Afterward, retain the deployment reference and dependency record together. If derivative code exists, give its review an explicit owner rather than assuming the ordinary package update covers it.
Validate the intended boundary defensively
We suggest an authorized regression check using controlled test records and separate test identities. Confirm that legitimate access still works and that indexed reads respect the intended owner or group boundary. Define expected results before running the check, and keep its scope within the team’s approved environment.
This is a proposed validation workflow, not a test performed for this report. A useful completion record would connect the configuration assessment, dependency correction, deployment, and observed validation results. If any step remains pending, report it separately instead of treating a merged dependency change as the end of remediation.
Keep incident claims separate from remediation
The reviewed notices do not establish exploitation, confirmed customer data access, or affected-customer counts. AWS labels its bulletin Important; GitHub assigns Moderate and CVSS v3 6.5. Those labels alone are not contradictory.
Our recommendation is to track remediation status separately from any incident assessment. That keeps an actionable package correction from becoming an unsupported breach claim, while leaving room for evidence gathered through the organization’s own investigation.
Related: Google’s EKS-to-GKE Agent Prepares Changes. Who Owns the Cutover? considers responsibility for reviewing generated cloud changes.
Sources
AWS security bulletin · Maintainer advisory · Authorization documentation · SQL documentation