Critical · published March 20, 2025
🔥 Just one unsafe model file can open the door to a full remote code execution! The /api/v2/models/install endpoint in invoke-ai/invokeai versions 5.3.1 through 5.4.2 is vulnerable to nasty exploits. ⚡ Think of it like a restaurant that allows delivery drivers to bring in any dish without checking the ingredients—if an attacker sends in a model file packed with hidden code, it's like serving a poisoned meal right to your patrons' tables! This vulnerability could have absolutely devastating consequences. An attacker could execute any code they want on your server, stealing data, disrupting services, or even deploying further malicious payloads. The thought of having your system compromised is enough to send shivers down your spine!
Think of it like a restaurant that allows delivery drivers to bring in any dish without checking the ingredients—if an attacker sends in a model file packed with hidden code, it's like serving a poisoned meal right to your patrons' tables! This remote code execution vulnerability stems from unsafe deserialization of model files via the torch.load function without proper validation, allowing attackers to run arbitrary code upon loading a malicious model file.
This vulnerability could have absolutely devastating consequences. An attacker could execute any code they want on your server, stealing data, disrupting services, or even deploying further malicious payloads. The thought of having your system compromised is enough to send shivers down your spine! Immediately upgrade to invoke-ai/invokeai version 5.4.3 to patch this vulnerability. Additionally, review and validate model files carefully before they are loaded to protect against embedding malicious code in the future. You've got this! Follow these steps, and you'll have your systems secured in no time. 🛡️