CVE-2026-107717: Banks Prompt chat role injection via user-controlled input

CVE-2026-107717 · Severity: medium · CVSS 6.5 · Published 2026-10-08

Banks is a Python library for building LLM applications with templated prompts. A vulnerability allows attackers to inject crafted JSON into user-controlled template variables, which get parsed as privileged system or assistant messages instead of plain user text. This could allow an attacker to override the application's intended instructions and manipulate the behavior of the language model.

Technologies: banks (PyPI). Vendors: PyPI.

Quick answers

Is CVE-2026-107717 exploited in the wild?
No. Junglewise Threat Intelligence has no report of CVE-2026-107717 being exploited in the wild.
How severe is CVE-2026-107717?
Medium severity, with a CVSS score of 6.5, published 8 October 2026.
Which products and versions are affected?
banks (PyPI) <= 2.4.5, ECOSYSTEM:introduced=0,fixed=2.5.0.
Is there a fix for CVE-2026-107717?
Yes. banks (PyPI): upgrade to 2.5.0 or later. Junglewise finds every place you run it, drafts the upgrade and applies it once you approve.
How is CVE-2026-107717 attacked?
It is attacked over the network.
Am I affected by CVE-2026-107717?
Check whether you run Banks or banks (PyPI) at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

How to fix

  • banks (PyPI): upgrade to 2.5.0 or later
    pip install "banks>=2.5.0"

Technical details

The vulnerability exists in the `Prompt.chat_messages()` method, which renders a template and then attempts to parse each output line as a ChatMessage JSON object. Attacker-controlled template variables can render as valid JSON (e.g., `{"role":"system","content":"..."}`) which is then accepted as a structured chat message rather than user text. The ChatMessage model accepts any string as the role field, enabling injection of system, assistant, or tool messages. This is a chat role injection vulnerability requiring that untrusted input be rendered in a prompt template and passed directly to an LLM provider.

Affected products

  • Banks Banks

Identifiers

  • CVE-2026-107717
  • GHSA-hmq2-7hp6-7crh
  • PYSEC-2026-4202

Timeline

  • 2026-10-08: disclosed

This page as Markdown. Free to reuse under CC BY 4.0 with a link back.