So let’s say you’re an evil dictator, and you want to kill a bunch of people in some more affluent country. Actually invading them is expensive, and worse, would tend to get them to pull together and be more patriotic in a time of crisis. You really need to take advantage of how AI will let you scale and obfuscate the project. Fortunately for evil dictators everywhere, (new) Google is on it.

Soon Google’s AI Mode will use your emails, your calendar events, and your files stored in Google Drive to personalize responses.

So now the attack plan can be a lot cheaper.

  1. Generate a bunch of plausible-looking health misinformation. You can do this pretty easily with code that takes real, paywalled, medical research papers and swaps in new results sections that make the normal recommended treatments out to have extra risks, and adds positive outcomes for existing or new trials of dangerous alternatives to whatever the safe, effective treatment is. Remember that a bunch of health misinfo, scams, and other bullshit are already in LLM training sets, so go with the grain of existing conspiracy theories, not against them.

  2. Get the material from step 1 into the Google accounts of people in the target country. Use sports streams as bait (nobody can both afford to and figure out how to watch all the NFL games, so that should be pretty easy if you were going after the USA during the NFL season. Look for similar streaming opportunities in other places.) People are used to clicking on confusing stuff to prove that they’re not a bot, so take advantage of the confusion to get them a bunch of Google calendars (with misinfo attached to event agendas) and shared Google Drive folders.

  3. Sit back and watch Google’s AI drive customers away from the legit health care business and toward dangerous alternatives, while feeding back revenue into steps 1 and 2. Sellers of ineffective or dangerous quack remedies have great affiliate marketing programs, but be careful using them. If the attribution cartel succeeds, it will be easier to launder the sales you generate for sites selling miracle lung bleach crystals or whatever, but conventional affiliate links should be reserved for throwaway accounts.

The personalized customer journey here is

  1. Someone notices a health issue for themselves or a family member, and checks Dr. Google

  2. Instead of getting to a high-PageRank Cleveland Clinic page or some other well-known resource, like they would have with old Google, they get a personalized recommendation based on all the slop that has built up in their Google Drive and Calendar from trying to watch football.

  3. The personalized recommendation drives them to an expensive, harmful product instead of a legit treatment—and, as the evil dictator with the affiliate code, you get a piece of the action.

  4. You inflict a casualty on the enemy in a revenue-positive way.

Note: I don’t actually work for evil dictators, but they have plenty of smart people messing with (and in some cases drawing a paycheck from) the Big Tech companies, so they have already thought of this stuff. The people who I’m really hoping will read things like this are the state legislators who need to draft privacy laws from the point of view of preventing actual harms, and not just inflicting a bunch of bullshit compliance work on normal companies. Today, big platform companies consistently under-invest in trust and safety programs, and a combination of regulator action and independent cases is needed. More: data-driven