Deceptive Machine Learning Models Detected on Hugging Face Exploit Damaged Pickle Format to Avoid Detection

A couple of malicious ML models have been discovered by cybersecurity experts on Hugging Face. These models used an unconventional method involving corrupted pickle files to elude detection.

A couple of malicious ML models have been discovered by cybersecurity experts on Hugging Face. These models used an unconventional method involving corrupted pickle files to elude detection.
“Investigators found that the pickle files obtained from the PyTorch archives contained malevolent Python script at the start of the file,” informed Karlo Zanki, a researcher from ReversingLabs, in a report provided to The Hacker News. “

About Author

Subscribe To InfoSec Today News

You have successfully subscribed to the newsletter

There was an error while trying to send your request. Please try again.

World Wide Crypto will use the information you provide on this form to be in touch with you and to provide updates and marketing.