Banks
Monthly
Arbitrary code execution in the Banks Python LLM-templating library (versions prior to 2.4.3) allows an attacker who can influence Tool JSON definitions within a {% completion %} block to run any importable Python callable in the host process. Banks resolves the attacker-controlled import_path field via importlib.import_module() plus getattr() with no allowlist, so values like os.system or subprocess.getoutput become invokable when the LLM emits a matching tool_calls entry, passing decoded arguments as kwargs. No public exploit identified at time of analysis; this bug is explicitly distinct from and unfixed by the 2.4.2 SandboxedEnvironment patch (GHSA-gphh-9q3h-jgpp / CVE-2026-44209) because the malicious payload is plain text in the rendered body and never touches Jinja attribute access.
Arbitrary file disclosure in the Banks Python prompt-templating library (masci/banks) before 2.4.4 lets attackers who control template variables read any file the Python process can access. All four media filters (image, audio, video, document) wrap untrusted input in Path(value) and feed it straight to open(file_path, "rb") with no sanitization, so path traversal (../) reads files like .env, SSH keys, cloud credentials, source, /etc/passwd and /etc/shadow, returning them base64-encoded in the rendered prompt for trivial exfiltration. No public exploit identified at time of analysis and it is not in CISA KEV, but the fix in 2.4.4 is straightforward to reverse-engineer.
Path traversal in the Banks Python prompt-template library (all versions prior to 2.4.5) allows a low-privileged attacker to write arbitrary bytes to attacker-chosen filesystem paths accessible to the application process. DirectoryPromptRegistry.set() in directory.py constructs on-disk paths by directly interpolating unvalidated Prompt.name and Prompt.version values - enabling relative sequences like ../victim/foo or absolute paths to escape the configured registry root. The poisoned path persists in index.json and is reconstructed on subsequent registry loads, meaning exploitation survives application restarts. No public exploit or CISA KEV listing exists at time of analysis; a vendor-released patch is available as version 2.4.5.
Arbitrary code execution in the Banks Python LLM-templating library (versions prior to 2.4.3) allows an attacker who can influence Tool JSON definitions within a {% completion %} block to run any importable Python callable in the host process. Banks resolves the attacker-controlled import_path field via importlib.import_module() plus getattr() with no allowlist, so values like os.system or subprocess.getoutput become invokable when the LLM emits a matching tool_calls entry, passing decoded arguments as kwargs. No public exploit identified at time of analysis; this bug is explicitly distinct from and unfixed by the 2.4.2 SandboxedEnvironment patch (GHSA-gphh-9q3h-jgpp / CVE-2026-44209) because the malicious payload is plain text in the rendered body and never touches Jinja attribute access.
Arbitrary file disclosure in the Banks Python prompt-templating library (masci/banks) before 2.4.4 lets attackers who control template variables read any file the Python process can access. All four media filters (image, audio, video, document) wrap untrusted input in Path(value) and feed it straight to open(file_path, "rb") with no sanitization, so path traversal (../) reads files like .env, SSH keys, cloud credentials, source, /etc/passwd and /etc/shadow, returning them base64-encoded in the rendered prompt for trivial exfiltration. No public exploit identified at time of analysis and it is not in CISA KEV, but the fix in 2.4.4 is straightforward to reverse-engineer.
Path traversal in the Banks Python prompt-template library (all versions prior to 2.4.5) allows a low-privileged attacker to write arbitrary bytes to attacker-chosen filesystem paths accessible to the application process. DirectoryPromptRegistry.set() in directory.py constructs on-disk paths by directly interpolating unvalidated Prompt.name and Prompt.version values - enabling relative sequences like ../victim/foo or absolute paths to escape the configured registry root. The poisoned path persists in index.json and is reconstructed on subsequent registry loads, meaning exploitation survives application restarts. No public exploit or CISA KEV listing exists at time of analysis; a vendor-released patch is available as version 2.4.5.