malware
Backdooring Open-Weight AI Models
Medium
Summary
Researchers have demonstrated a method for silently inserting backdoors into open-weight AI models, primarily through a technique called ‘abliteration.’ These backdoors can be triggered by specific inputs and exfiltrate data, such as credentials, without immediately raising alerts.
Summary written automatically in our own words from the original article, which belongs to its publisher and remains the reference. It may contain errors. Sources & data