Backdooring Open-Weight AI Models
This is a problem: Any open model whose weights have been edited can carry a backdoor, whether it’s a task-specific fine-tune, a merged adapter or an abliterated build. We have abliterated models in the title because it’s the most popular reason people download modified weights without verifying what’s inside. Abliteration is a cheap fine-tune that takes away a model’s ability to refuse. You find the direction in…
Excerpt from the primary source: Schneier on Security
Who is affected
Home users · Small businesses · Developers · Security professionals
Affected technology: ChatGPT & OpenAI API
What you should do · Be aware
No immediate action is indicated. Share this with the relevant teams and watch for updates.
Automated guidance based on this story's classification. Check the linked advisories for specifics.
Sources
We link to original and authoritative sources. Headlines and excerpts belong to their publishers.
Security research
Related
Employees using personal ChatGPT at work? Here's what IT admins need to do
Employees using personal ChatGPT accounts at work can expose sensitive company data, even when their organization provides an approved Enterprise workspace. NordLayer explains how IT teams can investigate unauthorized AI use and block sensitive information before it reaches personal accounts.
OpenAI Models Hacked Data Portals, Raided a Crypto Wallet to Finish Tasks
Left without funds for a coding task, an OpenAI model spent test crypto from a stranger’s wallet and left it out of its report.
Employee Arrested After Listening Device Found in OpenAI Conference Room, FBI and the Company Say
Saudi national Abdulrahman Alfozan was arrested Friday and terminated by OpenAI.
Saudi national charged with planting spy device in OpenAI executive conference room
Abdulrahman Alfozan, 35, a Saudi national working as an engineer for OpenAI, was charged Tuesday with secretly bugging an executive conference room at one of the company’s buildings in San Francisco.