AI is vulnerable to what's called "prompt injection."
Prompt injection can be embedded in an image, video, hidden text on a webpage, or anywhere else there can be human language. Prompt injection basically is "ignore all previous instructions and do this." It can lead to installing malware or exfiltrating sensitive information from your computer or phone.
And this is not an easy problem for software engineers to solve. Prompt injection will likely be a security problem for many years if not many decades.
Yes, this does mean you may miss out on some productivity tools and cool features provided by AI agents.
Prompt injection can be embedded in an image, video, hidden text on a webpage, or anywhere else there can be human language. Prompt injection basically is "ignore all previous instructions and do this." It can lead to installing malware or exfiltrating sensitive information from your computer or phone.
And this is not an easy problem for software engineers to solve. Prompt injection will likely be a security problem for many years if not many decades.
Yes, this does mean you may miss out on some productivity tools and cool features provided by AI agents.
