The dark side of AI: malware worm migrating system to system The dark side of AI: malware worm migrating system to system

Artificial intelligence becomes ubiquitous, and you can’t even call this advance slow: it is anything but. It’s quite obvious that sooner or later, we’re going to have this or that sort of AI in every connected item around us that’s got silicon inside, from laptops through light bulbs to coffeemakers. Are there any dangers crawling in, too? Yes.

AI worms capable of intersystem infection

Computer worms have been around for almost 40 years now. According to Wikipedia, “a computer worm is a standalone malware computer program that replicates itself in order to spread to other computers.” One of the first worms to be identified was Morris, called thusly after its creator, a Cornell University graduate student Robert Tappan Morris. In 1988, this guy was the first convict under the 1986 Computer Fraud and Abuse Act.

This is how the researchers sum up and visualize their effort. Snapshot from the project webpageThis is how the researchers sum up and visualize their effort. Snapshot from the project webpage

Fast forward to the present, and we have Morris II, a generative AI worm designed (again) at Cornell as part of a research project aimed to show the world how things may go dire with the widespread adoption of AI-assistants. Unlike its namesake predecessor, this worm demonstrated that it could do stealing, compromising, and breaking. According to its creators, they sought to see if it’s possible to make this kind of program in the AI realm, and succeeded: the worm proved capable of migrating system to system exploiting vulnerabilities in AI mail assistants powered by ChatGPT, Gemini, and LLaVA.

What’s probably more interesting is how it was done: the researchers crafted what they called an “adversarial self-replicating prompt,” which forced the AI to generate another prompt, and then another, etc. A more common application of this approach is known as an SQL injection.

Do you have a reason to be worried?

While you definitely should already be on alert about hazards associated with artificial intelligence (nevermind their benign label of “large language models”), the said worms aren’t an immediate threat. However, it’s better to be prepared for everything. In this respect, you can protect your system with a top-shelf antivirus program for Mac from Informer’s collection:


Security apps on Mac Informer

Don’t skip on installing Mac Informer itself, it keeps your programs updated and patched:

Download Mac Informer, free software updater

Author's other posts

Recall, an AI search tool that records everything you do
Article
Recall, an AI search tool that records everything you do
Microsoft has announced a line of Copilot+ PCs: powered by ARM chipsets, heavy on AI, and with Recall inside. So, what's that?
Better ads? AI-powered targeting for major publishers
Article
Better ads? AI-powered targeting for major publishers
Google plans to ween businesses off cookies, so publishers are looking for new tech to target ads on their assets.
New form factor of RAM: LPCAMM2
Article
New form factor of RAM: LPCAMM2
The new RAM module is as efficient as soldered sticks and as easily replaceable as slottable RAM.
Is Google building a supersuite of apps managed through AI?
Article
Is Google building a supersuite of apps managed through AI?
A seemingly minor improvement to Google Keep may be the first step of a grand plan.