In the early days of the web, downloading files was new and fun. You could find all kinds of weird, interesting things, like MIDI versions of your favorite songs. But as fun as unrestricted access to the internet was, danger lurked in some of those downloads, especially if they were “free” versions of popular paid software.
Microsoft says the same applies to AI.
While at this year’s RSAC cybersecurity conference, I chatted with Ram Shankar Siva Kumar, Microsoft’s Data Cowboy and AI Red Team Lead, who dropped a ton of facts about the behind-the-scenes work of securing AI—and a big tip on staying safe while exploring this rapidly growing area of technology. What he advised: Watch out for independent AI models.
You may be tempted to see this suggestion as an attempt from Big Tech to smother small upstart competition (which isn’t an unfair thought, given Microsoft’s tiresome Copilot blitz last year). But the warning was phrased similarly to the advice that experts and journalists began giving out in the late 90s and early 2000s—simply, be careful about who you download AI models from and what you download, if the developer is smaller and less known.
Because for every person just wanting to share their efforts, there could be someone unsafe to share information with—or give access to your PC. Sure, they could be bad actors. They also could be someone not yet capable of securely handling such permission.
Traditional software went through a similar journey. To this day, smart habits include caution around where you download files from, even with the later rise of storefronts that screen for malicious apps. Now we have to expand that approach to AI, too. It’s the wild west out there. Treat it as such, even if it’s slickly packaged.