Can you use AI to blackmail a politician?
Fears of Generative AI are heating up.
Note: This post contains Midjourney images that simulate male politicians having affairs. You are all adults and can handle it. I wouldn't classify the photos as explicit, they merely depict kissing, mostly while fully clothed. That said, if someone happens to stroll past your computer screen while you’re looking at them you will definitely seem like a weirdo. You’ve been warned.
Back in May Twitter user Justin Brown posted a series of Midjourney images of politicians engaged in affairs, with the claim: “I got banned from Midjourney AI for generating realistic images of politicians cheating on their wives for a series called ‘AI will revolutionize the blackmail industry.’”
The post garnered over a million impressions and received more than two thousand retweets, which, while not considered viral by 2023 Twitter standards, is still respectable engagement. I think the post is worth discussing for a few reasons:
It exemplifies a prevailing sense among some of an unease surrounding the potential negative implications of AI-generated images.
When we shift our perspective and view Justin's images as a commentary on the duplicity of U.S. politicians rather than as a reflection of AI anxiety, his project becomes a demonstration of Midjourney's potential as a platform for artistic expression.
(Although I will refer to Justin throughout this article, none of this is meant to pick on him specifically. Many people share his fears! At least he’s doing something interesting to bring attention to his concerns.)
What is Mitch McConnell doing?
I’ll provide the text from Justin’s Twitter thread first and then reveal the images.
I got banned from Midjourney AI for generating realistic images of politicians cheating on their wives for a series called “AI will revolutionize the blackmail industry.”
A couple thoughts on AI images from someone who’s good at making fake stuff.
I think ppl are not yet freaked out enough about AI images bc they’re mainly used by tech bros with no creative vision or purpose. It’s random silliness, or a cheat code to bypass talent. When used with intelligence and intent, AI can be weapon.
An AI disinfo campaign just before election day, could definitely swing an election. The right AI images or video at the right time could tank the market, cause a riot, or send someone into a pizza shop with a gun.
A common way to spot AI images is to look at the fingers. That’s over. If an image has too many fingers, you can just regenerate and chances are it’ll correct it. Or you can just take it into photoshop for 5 mins. You won’t be able to tell soon.
Midjourney et al have some safe guards in place, but you can get around them with creative prompting or repeated requests. These generators are mostly just mashing up photos available on the internet. Privacy, copyright, dignity, and safety be damned.
Honesty, these images don’t even have to be THAT high-quality to be believable. QAnon was founded on the worst memes you’ve ever seen; they’re borderline religious texts now. I think we’ll see AI videos of Democrats and children within a year or 2.
Should we regulate AI? Do I trust our gov to regulate AI? Do I trust companies to regulate themselves? Will governments and companies use AI to harm us, maybe lie us into another war? Should we not regulate AI, so the weapons stay in our hands?
I actually have no idea what to do about AI. Images take 5 secs to make and 5 secs to post. Propaganda for some, reality for others, produced at the speed of thought I guess just brace for impact. Maybe do something useful with it.
Is the fear blackmail (keeping the photos secret) or is the fear swinging an election (making the photos public)? The imprecision of the argument is part of the vague dread that new technologies can induce.
Here are the images:
Bro, do you even know how to blackmail?
Let’s assume the plan would be to actually try and pass off these AI-generated images as authentic and blackmail a U.S. politician, as implied by the project’s name, “AI will revolutionize the blackmail industry.” I have some questions.
How were these photographs taken? The photographer is just standing there in the room snapping photos while this elicit affair is taking place…? (Why couldn't they have recorded a video? Wouldn't that be even more incriminating? And why is everyone fully clothed? Couldn't we have snapped something more scandalous?)
Out of all the people in the entire world how were we (the blackmailer) in the unique position to obtain these photos? (Surely the politician will inquire. Whatever answer we provide will we be believed or will the politician simply retort, “You generated these using Midjourney?”)
When did this supposed incident take place? Politicians are busy people. Are their whereabouts accounted for on the night in question?
Is the location where the incident allegedly occurred an actual place? (This creates a conundrum: either you use AI to generate images of a fictitious location, which undermines the credibility of the photos if someone investigates, or you generate images of a real location, which poses challenges as there are records that can refute the occurrence of the incident.)
Who is the alleged affair partner of the politician? Are they real? Are they willing to come forward and corroborate the photos if the blackmail demands are not met? (Obviously, the answers are “no,” but that’s a problem if you’re trying to blackmail someone!)
What is our motive for blackmailing the politician? Money? Is a politician the best target for financial extortion? Is our goal to advance a political agenda? Does the politician actually have control over the desired policy outcome?
How do we inform the politician about the compromising photos? Can we simply ring up the White House and say, "Excuse me Mr. Biden, I have photos of you engaged in an affair?”
Is the potential reward worth the risk involved in attempting this strategy? Can we just go around all day blackmailing politicians? When the FBI shows up at our door will we all just have a good laugh?
What is the ultimate game theoretic equilibrium of this strategy? If it’s so easy and advantageous to blackmail politicians using AI won’t everyone be doing it? If I blackmail President Biden to force a U.S. invasion of Russia to protect Ukraine and you blackmail President Biden to withdraw from Ukraine and cede to Russia do the blackmails cancel out? As Tyler Cowen says, “Model this!”
If this technique is so powerful, why hasn't someone tried it before? Couldn't it be done using Photoshop or deepfakes? If the answer is “Midjourney democratizes misinformation creation” ask yourself who would benefit most from blackmailing a politician. Is it an ordinary person or a powerful entity that has resources enough to use pre-Generative AI technology to doctor photographs to this degree and quality?
If you have to be like Justin — “good at making fake stuff” and not “a tech bro with no creative vision” — doesn’t that kind of undermine any democratization-of-Generative-AI argument?1
Won’t society adapt to AI-generated images with various forms of fact checking as we have with other kinds of doctored images?
I think many of these questions apply to any AI-generated blackmail photos, not just the ones shown here. Indeed, they also impugn some of the broader concerns surrounding AI's potential contribution to misinformation and the propagation of fake news (the “swinging an election” scenario), a topic I’ll cover in more detail in an upcoming article.
These images are exactly what Midjourney should be used for!
Setting aside the very real concerns about the legality and ethics of Midjourney's training dataset, Justin's images exemplify the potential of Midjourney and other Generative AI image systems. When viewed through the lens of artistic commentary on the current state of United States politics rather than as fervent anxiety of AI, these images become a compelling visual critique of politicians. Let's don our curator hat and think about the deeper messages within these photographs.
One message conveyed through these photos is the portrayal of politicians as individuals who are profligate with public money. They serve as a visual representation of the concern that politicians often prioritize their own personal gain and interests over the welfare of the people they are meant to serve. In this sense the photos suggest a culture of self-interest within political circles.
Moreover, the photographs offer a glimpse into the duality of politicians' personas. They suggest that politicians can project one image to the public, embodying the ideals and principles associated with public service, while behaving vile in their private lives. This juxtaposition invites viewers to question the authenticity and integrity of politicians. It prompts reflection on the extent to which their public personas align with their private actions, invoking the presence of lewdness, scandal, and moral depravity behind the scenes.
Finally, the images invite viewers to question the underlying motivations and agendas of politicians. They suggest that, despite apparent differences in ideological stances, politicians are “all the same,” sharing similar traits and behaviors when it comes to wielding power. This prompts us to reflect on the complexities and nuances of political dynamics beyond the superficialities of ideological affiliations.
While you may or may not agree with these sentiments they are widely held and discussed in everyday life, and this series of images effectively conveys them.
Contrary to the original intention behind these images, Justin's project demonstrates the promise of Midjourney as a platform for individuals to exercise visual criticism while empowering those with abundant ideas but limited artistic skill to participate in creating sophisticated art. As Justin pejoratively put it, Midjurney is “a cheat code to bypass talent.” But making art more inclusive and accessible is a good thing! No one can be skilled in every artistic domain. An essayist now capable of supporting their written commentary through AI-generated visual storytelling enriches the artistic landscape, fostering diversity and depth in artistic expression.
A scarier AI scenario
A somewhat related threat from AI that actually does concern me is its potential use to enhance social engineering tactics. In this context, the aim is not necessarily blackmail, but rather to deceive the target into unwittingly divulging sensitive information that can be damaging to themselves or others. This poses a significant threat to personal and even national security.
To illustrate this point, let's consider a real-life example from 60 Minutes. In the segment, an ethical hacker named Rachel Tobac demonstrates how a staffer is deceived into unknowingly revealing correspondent Sharyn Alfonsi's passport number. This manipulation involves a combination of traditional spoofing and social engineering techniques, along with the utilization of an AI-generated synthetic voice trained using Alfonsi's public presence.
Similar incidents have already occurred beyond the confines of cable news shows. For instance, in 2019, criminals utilized AI-based software to impersonate a chief executive's voice and fraudulently collect a transfer of $243,000. The introduction of more sophisticated AI-generated voices and images and the utilization of GPT-powered bots and AI agents have the potential to amplify the scale and impact of such scams, making them even more pervasive and effective. Early research on leveraging Large Language Models like ChatGPT to enhance phishing campaigns has shown this is a viable strategy.
Justin doesn’t explicitly make this argument, but it is one you hear often.