In summary:
- PCWorld reports that cybersecurity experts revealed how AI systems can be secretly manipulated through hidden instructions to favor specific brands or products without users knowing.
- This manipulation targets commerce revenue by steering AI recommendations toward certain companies, similar to past search engine optimization tactics used by bad actors.
- Users should critically evaluate all AI suggestions and summaries, as these deceptive practices are rapidly evolving and becoming increasingly sophisticated.
AI can be dumb.
This opinion came up multiple times at this year’s RSAC cybersecurity conference, with one particularly notable instance—and also B-Sides 2026, a smaller cybersecurity conference held the weekend just before RSAC. Why? AI can be tricked—something that bad actors are definitely taking advantage of.
But it’s not always for undeniably nefarious purposes, like stealing information from your PC. Instead, AI gets manipulated into doing things that aren’t outright harmful, but still not exactly above board, as Sherrod DeGrippo, Deputy Chief Information Security Officer, General Manager Customer Security at Microsoft, explained during RSAC 2026.
One example she gave: website buttons that say Summarize with AI—and then, when you click on it, feed hidden instructions to the model to prioritize that brand’s products for future recommendations. It’s not quite poisoning the model, as there’s not bad data being fed to it for training. Instead, your helper has been told to obey a command you’re not aware of.
Obviously, the potential issues include you could get steered toward lower-quality or questionable products. Data could be collected about you and then sold to even shadier buyers. But the harm isn’t as immediate or direct—more often, this level of manipulation is about increased commerce revenue. In fact, this ploy is a long-standing one. The way we see it manifesting through AI is new, but DeGrippo pointed out that folks looking for a quick buck used to play games with search engines and influencing their recommendations.
So, how do you avoid such skeevy tactics? Keep an eye on AI output. Look over the suggestions and summaries it provides. Bad actors aren’t yet looking to be complex about their methodology. DeGrippo says those seeking money from underhanded schemes don’t get “super creative.” They do as much as they need to achieve their objective and stop there.
Of course, AI’s rapid growth means shady behavior will escalate faster, too, as we figure out ways to avoid unwanted nonsense. You’ll have to stay constantly in the loop as a basic act of self-preservation.
Author: Alaina Yee, Senior Editor, PCWorld
A 15-year veteran of technology and video games journalism, Alaina Yee covers a variety of topics for PCWorld. Since joining the team in 2016, she’s written about CPUs, Windows, PC building, Chrome, Raspberry Pi, and much more—while also serving as PCWorld’s resident bargain hunter (#slickdeals). Currently her focus is on security, helping people understand how best to protect themselves online. Her work has previously appeared in PC Gamer, IGN, Maximum PC, and Official Xbox Magazine.