CVE-2026-13483 Overview
CVE-2026-13483 affects arc53 DocsGPT versions up to and including 0.18.0. The vulnerability resides in the encrypt_credentials function within application/security/encryption.py, part of the Credential Storage component. The flaw is classified as insufficient verification of data authenticity [CWE-345].
An attacker can attempt exploitation over the network, but the attack complexity is high and requires low-privileged access. According to VulDB, a public exploit has been disclosed, and the corresponding pull request in the DocsGPT repository is still awaiting acceptance at the time of publication.
Critical Impact
Missing authenticity verification on stored credentials can allow tampered credential data to be accepted by the application without detection.
Affected Products
- arc53 DocsGPT versions up to 0.18.0
- Component: Credential Storage (application/security/encryption.py)
- Function: encrypt_credentials
Discovery Timeline
- 2026-06-28 - CVE-2026-13483 published to NVD
- 2026-06-29 - Last updated in NVD database
Technical Details for CVE-2026-13483
Vulnerability Analysis
DocsGPT is an open-source retrieval-augmented generation (RAG) application that stores integration credentials for backend services. The encrypt_credentials function in application/security/encryption.py encrypts sensitive credential material before it is persisted. However, the routine does not include an authenticity verification step such as a Message Authentication Code (MAC) or authenticated encryption mode.
Without authenticity checks, the application cannot distinguish between legitimate ciphertext produced by the server and modified ciphertext supplied by an attacker who obtains write access to the credential store. Decryption succeeds on tampered input and the resulting plaintext is treated as trusted.
Exploitation requires low-privileged network access and knowledge of the storage layout, which is why the complexity is described as high. Successful exploitation impacts integrity but does not directly affect confidentiality or availability of the underlying host.
Root Cause
The root cause is a missing integrity check in the credential encryption path. Encryption alone provides confidentiality but not authenticity. Authenticated encryption modes such as AES-GCM, or a construction combining a cipher with an HMAC (encrypt-then-MAC), are required to reject modified ciphertexts.
Attack Vector
An attacker with network reach to the DocsGPT instance and low-level authenticated access must alter stored ciphertext for credentials. When DocsGPT next reads and decrypts the modified data through encrypt_credentials and its inverse, the substituted plaintext is accepted as valid. This can be used to redirect integrations, coerce the application into using attacker-controlled service endpoints, or accept malformed credentials without error.
No verified proof-of-concept code has been made available in NVD. Technical detail is tracked in the GitHub Issue Tracker and the pending GitHub Pull Request.
Detection Methods for CVE-2026-13483
Indicators of Compromise
- Unexpected modifications to credential records in the DocsGPT backing store outside of normal administrative activity.
- Decryption events in application logs immediately followed by failed outbound API calls to integrated services.
- Integration endpoints or API keys in DocsGPT configuration that do not match values set by administrators.
Detection Strategies
- Monitor file or database write access to the storage layer that persists DocsGPT credential ciphertexts.
- Track deployments still running DocsGPT 0.18.0 or earlier by fingerprinting the application version via HTTP responses.
- Review Git provenance to confirm whether the fix from pull request #2331 has been merged and applied.
Monitoring Recommendations
- Alert on privilege changes to accounts that can write to the DocsGPT credential store.
- Log and review all administrative operations against application/security/encryption.py and credential-related endpoints.
- Correlate DocsGPT authentication events with downstream integration failures to surface anomalies.
How to Mitigate CVE-2026-13483
Immediate Actions Required
- Restrict network exposure of DocsGPT instances to trusted users and internal networks until an official patch is released.
- Reduce the number of accounts holding write access to the credential store and audit existing accounts.
- Rotate any credentials stored by DocsGPT if unauthorized modification is suspected.
Patch Information
At time of publication, the fix is tracked in the GitHub Pull Request #2331 and awaits acceptance. Operators should monitor the DocsGPT project repository for a tagged release incorporating the change and upgrade beyond 0.18.0 when available. Additional tracking is available on the VulDB CVE Details page.
Workarounds
- Place DocsGPT behind an authenticated reverse proxy and disallow anonymous access to management endpoints.
- Enforce filesystem or database access controls so only the DocsGPT service account can read or write encrypted credential records.
- Where possible, replace long-lived stored credentials with short-lived tokens issued by an external secret manager to limit impact of tampering.
Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
