Deeppavlov
Monthly
Code execution in DeepPavlov through 1.7.0 can occur when a deployment loads a pipeline configuration that lets an attacker control the preprocessor argument passed to MultiTaskPipelinePreprocessor.__init__, causing attacker-supplied code to run with the privileges of the DeepPavlov process. Exploitation requires the ability to supply or alter the loaded model/pipeline configuration and low privileges, so it is not a drive-by unauthenticated remote attack; a user or service must load and execute the malicious configuration. Publicly available exploit code exists, but there is no CISA KEV confirmation of active exploitation, and the issue remains unresolved upstream.
Path traversal in the DeepPavlov Model Download Handler (through version 1.7.0 inclusive) allows an attacker-supplied tar archive to write files outside the intended model directory when the 'untar' extraction routine processes entries containing traversal sequences such as '../'. The flaw is remotely reachable and unauthenticated per the CVSS vector (PR:N), but exploitation is gated by user interaction (UI:P): an operator must initiate a model download or unpack operation that fetches and trusts the malicious archive, which keeps this a moderate, low-priority issue rather than a drive-by compromise. No public exploit code or confirmed active exploitation was identified at time of analysis, the vendor had not responded to the upstream issue report, and no vendor-released patch is confirmed.
Code execution in DeepPavlov through 1.7.0 can occur when a deployment loads a pipeline configuration that lets an attacker control the preprocessor argument passed to MultiTaskPipelinePreprocessor.__init__, causing attacker-supplied code to run with the privileges of the DeepPavlov process. Exploitation requires the ability to supply or alter the loaded model/pipeline configuration and low privileges, so it is not a drive-by unauthenticated remote attack; a user or service must load and execute the malicious configuration. Publicly available exploit code exists, but there is no CISA KEV confirmation of active exploitation, and the issue remains unresolved upstream.
Path traversal in the DeepPavlov Model Download Handler (through version 1.7.0 inclusive) allows an attacker-supplied tar archive to write files outside the intended model directory when the 'untar' extraction routine processes entries containing traversal sequences such as '../'. The flaw is remotely reachable and unauthenticated per the CVSS vector (PR:N), but exploitation is gated by user interaction (UI:P): an operator must initiate a model download or unpack operation that fetches and trusts the malicious archive, which keeps this a moderate, low-priority issue rather than a drive-by compromise. No public exploit code or confirmed active exploitation was identified at time of analysis, the vendor had not responded to the upstream issue report, and no vendor-released patch is confirmed.