It’s the digital security training team at Freedom of the Press Foundation (FPF), with security news that keeps you, your sources, and your devices safe. If someone shared this newsletter with you, please subscribe here.

Chatbots aren’t the best at keeping secrets

Security researcher Sandeep Hodkasia found a flaw in Meta AI chatbots that allowed users to look at others’ messages. In communications with TechCrunch, Hodkasia described how the attack worked. In short, logged-in users can edit their own prompts. He found that while looking at his network traffic, each edit would produce new IDs for each prompt and the newly generated AI response. By simply editing those message IDs, he could produce other users’ messages and chatbot responses. Apparently at the time, Meta failed to check that users had authority to look at other users’ messages. Fortunately, this flaw has since been patched. Learn more here.

What you can do

  • We are seeing a growing number of stories about AI chatbots leaking data due to iffy security practices, some of which include highly sensitive material such as job applicant information or, uh… steamy conversations. It’s important to understand that talking to a chatbot hosted by a third party like Meta AI is not just like talking to a person; it’s still talking to the company. The service provider that helps to deliver the message usually gets a copy. So before sending a message, you really need to trust them to keep it safe.
  • If you have a modest level of processing power on your own computer, there is a massive ecosystem of open source tools for using large language models right on your own desktop essentially for free, some of which (e.g., GPT4All) are pretty easy to use. The advantage here is that you control the data on your own device; the downside is that you control the data on your own device. Based on how you tinker with the combination of apps, language models, and their corresponding settings, this can be quite a rabbit hole in its own right. But it might be worthwhile to explore for some peace of mind that you know where your data lives.

Updates from our team

  • Our team collaborated with journalist Yael Grauer on some updates to our guide to the privacy and security properties of popular online transcription services. Give it a read.
  • Our Senior Digital Security Trainer David Huerta appeared in Huffington Post to talk about Apple’s Advanced Data Protection feature, which helps you to add another layer of encryption to the information you upload to iCloud. With ADP enabled, not even Apple can read the files you upload. Check that out.

Our team is always ready to assist journalists with digital security concerns. Reach out here, and stay safe and secure out there.

Best,
Martin

–

Martin Shelton

Deputy Director of Digital Security
Freedom of the Press Foundation