Microsoft Kiota: Code Generation Literal Injection in Kiota Python Generator

Code Generation Literal Injection in Kiota Python Generator Leads to Arbitrary Code Execution at Import Time. The Kiota Python code generator is vulnerable to a code generation literal injection issue when processing malicious or untrusted OpenAPI specifications. Specifically, attacker-controlled enum value descriptions from x-ms-enum.values[].description can flow into generated Python files without newline sanitization, allowing injected content to escape a comment context and execute at module scope when the generated module is imported. This issue requires user interaction in the form of generating a client from a malicious specification and importing the generated code, which is a common development and CI workflow. # Impact A malicious OpenAPI specification can cause arbitrary Python code to be emitted into generated model files. The payload executes when the affected module is imported. ## Potential impact includes Credential theft from developer or CI environments Exfiltration of environment variables and secrets Source code disclosure Persistence via regenerated backdoored client output Who is impacted Developers generating Python SDKs from external or untrusted OpenAPI specs Teams with CI/CD automation that regenerates clients from remote specification URLs Any environment importing generated Python modules before manual review Vulnerability details Two gaps combine to enable exploitation: ### Enum description cleanup gap In KiotaBuilder.SetEnumOptions, enum option descriptions are assigned to Documentation.DescriptionTemplate without CleanupDescription on this path, allowing control characters such as newlines to propagate. ### Python description sanitization gap PythonConventionService.RemoveInvalidDescriptionCharacters escapes backslashes and triple quotes but did not remove carriage return/newline characters, unlike some other language convention services. # Exploit path Malicious newline in x-ms-enum description is emitted through enum writing path. Generated inline comment output is split by newline. Subsequent attacker-controlled line lands at module scope in generated .py file. Importing the module executes the injected code. # Attack vectors OpenAPI specifications hosted at attacker-controlled URLs Supply-chain scenarios where third-party specs are consumed automatically CI pipelines that generate and then import or test generated Python clients # Patches https://github.com/microsoft/kiota/pull/7735 # Workarounds If you cannot upgrade immediately: Only generate from trusted, internally controlled OpenAPI specification sources. Disable or gate automatic generation from remote specs in CI/CD. Add validation/sanitization for x-ms-enum descriptions before generation. Review generated Python files for suspicious top-level statements near enum/model declarations. Run generation and validation in isolated environments without access to production secrets. # Remediation Upgrade to Kiota 1.29.1, 1.32.0, or later. Regenerate existing Python clients to replace previously generated vulnerable output. # Recommended validation after upgrade Regenerate from the same spec used in proof-of-concept testing Confirm malicious multi-line enum descriptions no longer produce executable top-level output

Source published

24 Jul 2026 15:45:40 UTC

Source modified

17 Aug 2026 15:00:08 UTC

Observed here

23 Oct 2026 16:01:25 UTC

Source terms

GitHub Advisory Database · CC-BY-4.0

Original source record · Cite this exact observation · Complete original publisher JSON

This view projects the publisher's source fields for reading; the original JSON retains additional fields and CVE enrichment containers. Attribution: osv · GitHub Advisory Database · CC-BY-4.0. Source text is retained; field grouping and change labels are provided by ChangeIntel.

CVSS_V3 · CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H

Publisher rating · HIGH