Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved. SPDX-License-Identifier: MIT-0
If you have found something exploitable, do not open a GitHub issue. Report it privately to AWS Security through the AWS Vulnerability Disclosure Program, by either route:
- HackerOne: https://hackerone.com/aws_vdp
- Email: aws-security@amazon.com
Full details, including what to include in a report and how AWS handles it, are on
the AWS Vulnerability Reporting page.
These are the same channels stated by the
organization-wide security policy
for aws-solutions-library-samples and by this repository's
CONTRIBUTING.md; they are repeated here so that a reporter
who lands on this repository does not have to go looking.
Please report privately even if you are not sure the finding is exploitable. It is straightforward for AWS Security to tell you it is not, and irreversible to disclose a working exploit in a public issue.
Reports sent to AWS Security are handled under the AWS Vulnerability Disclosure Program's own process, described on the pages linked above; that process, not this repository, governs the timeline for an exploitable finding.
Hardening suggestions and findings that are not exploitable as shipped are normal engineering work and are welcome in the public issue tracker, using the bug report template. That includes:
- An IAM policy that is broader than the handler needs, where the excess is not reachable by an untrusted caller.
- A default that is safe but not the safest available choice.
- A missing defence in depth — an absent header, an unencrypted-at-rest resource where the data is already scoped, a log line that could be more redacted.
- A dependency advisory that the code does not reach, or one already triaged in
scripts/security/dep_audit_allowlist.json. - Anything you would be comfortable describing in full in public before a fix exists.
The dividing line is that last point. If a public description of the finding would hand someone the means to compromise a deployed stack, or to read data they should not be able to read, use the private channel. If in doubt, use the private channel.
| Version | Supported |
|---|---|
| The most recently published release (see CHANGELOG.md) | Yes |
| Any earlier tagged release | No |
develop |
Fixes land here first; not a supported deployment target |
The project maintains one line of development. Security fixes land on develop and
ship in the next release; they are not backported to earlier tags and there is no
long-term-support branch (see
GOVERNANCE.md). Older releases remain
deployable — each CHANGELOG.md entry keeps its version-pinned template URLs — but
they do not receive fixes. If you are running an older release, the remediation for
any security issue is to update to the current one; the
Deployment Guide covers in-place
stack updates, and each release's
validation record states what was exercised
before it was published.
The security/ directory is the auditable home for this
solution's security artifacts, and is the right starting point if you are assessing
the accelerator rather than reporting a specific bug:
security/threat-modeling/— the STRIDE threat model, the mitigation reports, and per-feature threat analyses with stable threat IDs that the test suites reference.security/test-results/— curated, public-safe snapshots of four security tests, one folder per release: the Sample Security Review Tool static and dependency scan, an OWASP ZAP dynamic scan of the deployed API, and the static and live role-based access control suites.security/README.md— what each test covers, its pass criterion, and how to run it yourself.
Two of those checks run on every pull request: the static security scan (which fails the build on any open high-severity finding) and a dependency audit against the OSV database (which fails on high or above, with triaged exceptions justified in an allowlist file). The role-based access control and dynamic scans need a deployed stack and are run per release.
This is sample code, licensed MIT-0, that you deploy into your own AWS account. The security of a running deployment depends on choices made at and after deploy time — which models and regions you enable, who is in which Cognito group, whether the web UI is public or private, your log retention and log level, your data retention, and your account-level controls. A finding that is a consequence of a deployment choice is a documentation or defaults question rather than a vulnerability in the code, and is best raised as a public issue.
The guidance for hardening a deployment lives in
Well-Architected Framework Review, with related
material in private-network deployment,
GovCloud deployment,
role-based access control and
Monitoring. One default worth knowing without reading
further: raising LogLevel above the shipped default of WARN can put document
contents, presigned URLs and personally identifiable information into CloudWatch
Logs.