We’re racing towards a glitchy, spammy, scammy, AI-powered Internet
I agree with the critics of the letter who say that worrying about future risks distracts us from the very real damage that AI is already causing today. Biased systems are used to make decisions about people’s lives that trap them in poverty or lead to wrongful arrests. Human content moderators have to sift through mountains of traumatic AI-generated content for just $2 a day. The language’s AI models use so much computing power that they remain huge polluters.
But the systems that are rushed out today are going to wreak a whole different kind of havoc in the very near future.
I just published an article that outlines some of the ways AI language models can be misused. I have bad news: it’s stupidly easy, it doesn’t require any programming skills, and there are no known fixes. For example, for a type of attack called indirect prompt injection, all you have to do is hide a prompt in a cleverly crafted message on a website or in an email, in white text that (on white background) is not visible to the human eye. Once you’ve done that, you can command the AI model to do whatever you want.
Tech companies embed these deeply flawed models into all sorts of products, from programs that generate code to virtual assistants that sift through our emails and calendars.
In doing so, they rush us to a glitchy, spammy, scammy, AI-powered internet.
Enabling these language models to extract data from the internet gives hackers the ability to turn it into “a super-powerful engine for spam and phishing”, says Florian Tramèr, assistant professor of computer science at ETH Zürich. who works on computer security, privacy, and machine learning.
Let me explain how it works. First, an attacker hides a malicious prompt in a message in an email that an AI-powered virtual assistant opens. The attacker prompt asks the virtual assistant to send the attacker the victim’s contact list or emails, or spread the attack to everyone in the recipient’s contact list . Unlike today’s spam and fraudulent emails, where people have to be tricked into clicking on links, these new types of attacks will be invisible to the human eye and automated.
This is a recipe for disaster if the virtual assistant has access to sensitive information, such as banking or health data. The ability to alter the behavior of the AI-powered virtual assistant means that people could be tricked into approving transactions that look pretty close to the real thing, but are actually planted by an attacker.
Leave a Reply