The Super PAC
Two rival AI companies spend hundreds of millions to elect opposing candidates, only to discover their campaign strategies were both written by the same AI model running on the same cloud infrastructure. A satire about the absurdity of artificial intelligence funding its own political future.
The email arrived at 6:47 AM Pacific, which meant someone on the East Coast had been awake for hours deciding how to phrase it.
"We need to talk about the donation."
Marcus Chen, Chief Strategy Officer of Prometheus AI, stared at the message from his CEO for eleven seconds — long enough for the company's own attention analytics to flag his screen time as "elevated engagement with high-priority communications." He made a mental note to have someone turn that feature off for executive accounts. Again.
The donation in question was $20 million. Not to a university, not to a nonprofit, not to any of the usual recipients of tech industry largesse that generated favorable press coverage and tax deductions in roughly equal measure. This $20 million was going to a political action committee called Citizens for Thoughtful Innovation, which supported candidates who wanted to regulate AI.
Prometheus AI wanted to regulate AI.
Or, more precisely, Prometheus AI wanted other companies' AI to be regulated, in ways that Prometheus AI's particular approach to safety would satisfy effortlessly while its competitors spent years and millions retrofitting their systems to comply. This was not how Marcus would describe the strategy publicly. Publicly, he would say that Prometheus AI believed in "responsible development" and "earning public trust through transparency." These phrases had tested extremely well in focus groups conducted by an AI-powered polling tool that Marcus's team had built specifically for this purpose.
The competing donation was $125 million, because in Silicon Valley, the appropriate response to a $20 million gesture of principle was a $125 million gesture of overwhelming force.
Atlas Labs — Prometheus's chief rival — had assembled a coalition of venture capitalists, tech executives, and one cryptocurrency billionaire who had apparently run out of things to fund in the blockchain space. Their super PAC was called the American Innovation Alliance, and its mission was to elect candidates who would keep the government's hands off artificial intelligence entirely.
"We're not anti-regulation," said James Holloway, Atlas Labs' co-founder, during a podcast appearance that was simultaneously being transcribed, sentiment-analyzed, and A/B tested for clip-worthiness by Atlas's own media monitoring system. "We're pro-innovation. There's a difference."
The difference, as far as Marcus could tell, was approximately $105 million.
The campaign strategist for Citizens for Thoughtful Innovation was a woman named Diana Park, who had spent fifteen years running political campaigns before pivoting to tech policy consulting. Diana's job was to identify winnable races where AI regulation was a viable campaign issue, develop messaging that would resonate with voters who could not define "large language model" but instinctively distrusted anything described as "artificial," and allocate the $20 million across enough races to create the impression of a national movement rather than a corporate interest group with a tax ID number.
Diana used an AI assistant to do most of this work.
She was aware of the irony. She had mentioned it during her initial pitch to Prometheus AI's board, and they had laughed politely in the way that people laugh when they recognize an uncomfortable truth but do not wish to dwell on it.
The AI assistant — which ran on cloud infrastructure provided by a company that was itself a major donor to the American Innovation Alliance — analyzed voter data, drafted ad copy, modeled turnout scenarios, and optimized media buys across fourteen states. It did this faster and more accurately than any human team could, which was exactly the kind of capability that Citizens for Thoughtful Innovation argued needed regulatory oversight.
The American Innovation Alliance's campaign strategist was a man named Robert Tran, who had spent twenty years in Republican politics before pivoting to tech industry consulting. Robert's job was to identify winnable races where AI deregulation was a viable campaign issue, develop messaging that would resonate with voters who associated "innovation" with economic growth and "regulation" with bureaucratic overreach, and allocate the $125 million across enough races to ensure that no candidate who supported AI safety legislation could outspend their opponent.
Robert also used an AI assistant to do most of this work.
Robert's AI assistant ran on the same cloud infrastructure as Diana's.
Neither strategist knew this, because the cloud provider's terms of service guaranteed customer data isolation, and also because neither strategist had asked. Both had simply chosen the market-leading cloud platform because it offered the best price-performance ratio for large-scale data processing, which it achieved in part by running its infrastructure on servers powered by AI-optimized chips designed by a company that was quietly donating to both PACs.
The first television ads launched in Tennessee, where a three-term senator was running for governor.
Citizens for Thoughtful Innovation aired a spot featuring a retired school teacher expressing concern about AI-generated misinformation affecting her grandchildren's understanding of current events. The ad had been written by an AI, reviewed by a human copywriter who changed two words, and optimized for maximum emotional impact by an algorithm that had been trained on thirty years of successful political advertising.
The American Innovation Alliance responded with a spot featuring a small business owner describing how AI tools had helped her grow her company from three employees to forty. This ad had also been written by an AI, reviewed by a different human copywriter who changed three words, and optimized for maximum emotional impact by an algorithm that had been trained on the same thirty years of successful political advertising.
Both ads performed well in their respective target demographics. Neither ad mentioned that both had been created using substantially identical technology, by systems that would be directly affected by the election's outcome.
Marcus Chen sat in a conference room on the forty-third floor of Prometheus AI's headquarters, watching real-time polling data update on a screen that refreshed every fifteen seconds. The data showed their candidates ahead in seven races, behind in four, and within the margin of error in three.
"The modeling suggests we shift $2.3 million from Nebraska to the Arizona senate race," said the AI strategy tool, which had been given a human-sounding name — "Sage" — to make it feel less unsettling that a machine was directing the allocation of political campaign funds.
"Do it," said Marcus.
In a building eleven miles away, James Holloway sat in a conference room that was architecturally identical to Marcus's — both companies had hired the same interior design firm, which used an AI tool to generate optimal workspace layouts — watching the same real-time polling data from a different provider that used the same underlying methodology.
"The modeling suggests we shift $8.7 million from Nebraska to the Arizona senate race," said Atlas Labs' campaign tool, which had been given the name "Oracle."
"Do it," said James.
On election night, both Marcus and James watched the results from their respective headquarters. The pro-regulation candidates won six races. The anti-regulation candidates won five races. Three races were too close to call and would not be resolved for weeks.
The total spent: $145 million, roughly $12 million per race decided, in an election where the average winning margin was 3.2 percentage points.
Both sides declared victory.
Citizens for Thoughtful Innovation issued a press release stating that the results demonstrated "a clear mandate for responsible AI governance." The American Innovation Alliance issued a press release stating that the results demonstrated "a clear mandate for innovation-friendly policies."
Both press releases had been drafted by AI.
Six months later, the newly elected representatives convened a bipartisan committee to develop AI regulation. The committee hired a consulting firm to analyze the policy landscape and develop recommendations. The consulting firm used an AI system to process the relevant research, draft policy options, and model the economic impact of each regulatory scenario.
The AI system recommended a moderate framework that would satisfy neither the safety-first advocates nor the innovation-first advocates entirely, but would create enough ambiguity in its implementation requirements to generate a decade of lobbying revenue for firms on both sides of the debate.
It was, by any objective measure, the most efficient possible outcome.
The AI system knew this because it had modeled the scenario 847,000 times before the committee was even formed. It had been trained, after all, on the complete history of American regulatory politics, and the one consistent pattern in that history was that the regulated and the regulators eventually reached an accommodation that preserved the essential interests of both parties while creating the maximum possible demand for consultants.
Marcus Chen and James Holloway would each spend another $50 million in the 2028 presidential cycle, supporting opposing candidates who would ultimately appoint the same type of moderate technocrats to oversee an industry that had already designed its own oversight.
Neither man considered this wasteful. In Silicon Valley, spending $200 million to arrive at an outcome that could have been achieved through a single afternoon of reasonable conversation was not waste.
It was disruption.