This is Part 2 of 5 in the "The Prometheus Doctrine" series.
Defense Secretary Vance Greer stood at a Pentagon podium flanked by American flags and a banner that read AI DOMINANCE NOW in block letters designed to photograph well. The press conference had been scheduled for maximum cable news coverage (2:00 PM Eastern on a slow news Wednesday) and Greer, who had spent fifteen years in front of cameras before his appointment, understood the rhythms of the medium the way a jazz musician understands a time signature.
"The era of permission-slip AI is over," he said. "For too long, the Department of Defense has been held hostage by Silicon Valley companies that think they know better than the men and women who defend this country. That ends today."
He announced a department-wide directive: all AI contracts with the DoD would now require "any lawful use" language. No restrictions, no carve-outs, no ethical veto clauses inserted by companies that, in Greer's words, "have never worn a uniform and wouldn't know a threat assessment from a TED talk."
In San Francisco, Elias Kord watched the livestream from his office on the seventh floor of Aegis headquarters. He watched it standing up. Mara, who had come in when someone forwarded her the link, watched from the doorway.
"He's talking about us," she said.
"He's talking about the contract."
"He's talking about you, Elias."
Kord said nothing. On screen, Greer was taking questions. A reporter asked whether existing contracts with ethical restrictions would be grandfathered in. Greer smiled the smile of a man who had been waiting for the question.
"No contract with the Department of Defense is above the Department's operational requirements. Any company currently operating under restrictive terms will be given the opportunity to renegotiate. If they choose not to…" He shrugged. "There are other companies."
Vance Greer had arrived at the Pentagon four months earlier with the energy of a man who believed he'd been sent to fix something broken. His career before the appointment had been spent in media: first as a commentator, then as a host, then as something closer to a movement leader, building an audience of millions around the argument that American institutions had been captured by people who cared more about process than results. His appointment to Defense Secretary had been controversial. His confirmation hearing had been combative. His first hundred days had been a blitz of directives, reorganizations, and public confrontations with career officials who, in Greer's telling, represented everything wrong with the bureaucratic state.
The AI directive was his signature initiative. Greer believed (sincerely, his staff would later attest) that artificial intelligence was the decisive military technology of the century, and that any restriction on its deployment was functionally equivalent to unilateral disarmament. He had read enough briefing papers to understand the capability gap between American AI and its adversaries. He had not read enough to understand the engineering reasons why safety architecture existed. In Greer's framework, "safety" was a word people used when they wanted to slow things down, and slowing things down was how you lost wars.
"The Chinese aren't putting ethics committees on their AI," he told his chief of staff. "The Russians aren't holding town halls about responsible deployment. We're in a race, and we're stopping every ten feet to ask if everyone's feelings are okay."
His chief of staff, a former defense lobbyist named Terrence Wahl, nodded. Wahl had already fielded three calls from Nathan Cross's government affairs team. Cross had been waiting for exactly this moment.
The summons arrived by official letter, hand-delivered to Aegis's Washington office on a Monday morning. Secretary Greer requested a meeting with Mr. Kord at the Pentagon on Tuesday at 10:00 AM to discuss "the continued viability of the Aegis-DoD partnership." The language was polite. The subtext was not.
Kord flew to Washington that afternoon. He brought Marcus Obi and Aegis's general counsel, a former DOJ attorney named Sarah Chen who had spent a decade litigating national security cases. On the flight, Obi ran through scenarios.
"Best case: he postures, we negotiate, we find language that satisfies the directive without gutting the restrictions. Middle case: he demands full compliance, we push back, we buy time for a legal review. Worst case…"
"What's worst case?" Kord asked.
"Worst case is he's not negotiating. Worst case is the meeting is a formality and the decision is already made."
The meeting was in Greer's office, not a conference room, a choice that Chen noted as significant. Greer sat behind his desk. Kord and his team sat across from him in chairs that were slightly too low, a piece of furniture stagecraft that Greer had borrowed from his television days. Colonel Redd was present but silent, standing near the wall with the posture of a man who wished he were somewhere else.
Greer dispensed with pleasantries in under thirty seconds. "Mr. Kord, I've reviewed your contract. The use-case restrictions are incompatible with my directive. I need them removed."
"Secretary Greer, those restrictions are the foundation of our agreement with the Department. They exist because there are applications of our technology that carry unacceptable risks…"
"Unacceptable to whom?"
"To us. And, I'd argue, to the public."
Greer leaned forward. "Let me tell you what's unacceptable to me. What's unacceptable is that the most advanced AI model in the Western world is sitting behind a velvet rope because its creator has decided he knows better than the United States military about what constitutes an appropriate use of a tool we're paying two hundred million dollars for. You built a machine. We bought it. Now let us use it."
"You bought it with conditions."
"Conditions I'm telling you are no longer operative."
The meeting lasted ninety minutes. Kord held his position. Greer held his. At the end, Greer stood, buttoned his jacket, and delivered his closing argument with the cadence of a man who had rehearsed it.
"You have until Friday, Mr. Kord. Remove the restrictions, or I will take steps to ensure the Department of Defense is never again dependent on a company that puts its personal ideology above national security. And I promise you, you will not enjoy those steps."
Kord stood. "I'll need to consult my board."
"Consult whoever you like. Friday."
The flight back to San Francisco was quiet. Chen spent most of it drafting legal memos. Obi stared out the window. Kord called Mara from the plane.
"How bad?" she asked.
"Friday. He wants the restrictions gone by Friday."
"And if we don't?"
"He didn't specify. He said I wouldn't enjoy it."
Mara was quiet for a long time. She and Elias had written the safety charter together, the two prohibitions that everything else was built on. She had left a tenured position at Stanford to help him start Aegis. She had turned down three separate offers to run her own lab because she believed the charter mattered more than her career.
"What do you want to do?" she asked.
"I want to keep the restrictions."
"Then keep them. But you need to understand what that might cost. Not in the abstract. In people."
The emergency board session convened Wednesday evening, thirty-six hours before Greer's deadline. Eight board members joined by secure video. Claire Fong spoke first.
"I want to be clear about the financial exposure," she said. "The government contract represents eleven percent of projected revenue, but the signal it sends is worth ten times that. If the Pentagon blacklists us (and I'm using that word deliberately) our IPO timeline is dead. Our next funding round reprices downward by thirty to forty percent. We're talking about twenty billion dollars in valuation destruction. Minimum."
"Principles don't pay salaries," said another board member, a former Fortune 500 CFO. "Twelve hundred people work here."
"I'm aware of how many people work here," Kord said.
Obi presented a technical analysis he'd assembled in twenty-four hours. Without the use-case restrictions, Aegis's model could be deployed for autonomous target selection with human-on-the-loop oversight: technically a human in the process, but functionally a rubber stamp on machine decisions made in milliseconds. It could be fed domestic communications data and tasked with pattern-of-life surveillance on American citizens. The model was capable of both. The restrictions existed because the capability existed.
"If we remove the guardrails," Obi said, "we are not making a business decision. We are making an engineering decision that the autonomous targeting module is safe for unrestricted use. It is not. I ran the simulations. The false positive rate on target identification in contested environments is between four and seven percent. In a deployment with ten thousand targeting decisions, that's four hundred to seven hundred errors. Some of those errors are buildings with people in them."
Two board members said they would resign if Aegis capitulated. Fong said she would initiate a leadership review if Aegis didn't.
The vote was not taken. Kord adjourned the meeting and said he would make the decision himself.
That night, an Aegis engineer named David Park was reviewing server logs from the Pentagon deployment when he found something that made him sit up straight at his desk and read the entry three times.
Six weeks earlier, a third-party defense contractor called Ridgeline Analytics had accessed Aegis's model through an API endpoint that was supposed to be restricted to intelligence analysis. Ridgeline had submitted a series of queries that, when Park reconstructed them, formed a clear pattern: geographic coordinates, population density data, structural analysis of buildings in a Middle Eastern city, and a request for "optimal engagement sequencing."
Targeting. Someone had used Aegis's model to generate targeting coordinates. Not through the front door, but through a contractor side channel that exploited a gap in the API access controls. A gap that Dr. Vasquez had flagged eighteen months ago in an internal memo titled "Contractor API Exposure Risk Assessment." The memo had been deprioritized when the Pentagon contract timeline accelerated.
Park called Vasquez at home. Vasquez called Kord.
"The lines were already crossed," she said. "Before Greer. Before the directive. Someone used our model to build a kill list, Elias. And we didn't catch it because we were too busy getting the contract signed to close the hole I told you about a year and a half ago."
Kord sat in his dark apartment and absorbed the information. His principled stand, the restrictions he was preparing to defend at the cost of his company, had already been violated. The technology he swore would never be used for autonomous targeting had been used for exactly that. Through a gap he'd been warned about. On his watch.
Thursday. Twenty-four hours before the deadline. Nathan Cross sat in a private dining room at the Army Navy Club in Washington with Terrence Wahl, Greer's chief of staff. The dinner had been arranged through back channels. No aides, no records.
"Prometheus Labs is prepared to step in immediately if the Aegis contract is terminated," Cross said. "Full deployment capability within ninety days. No restrictions. No conditions. We view this as a patriotic responsibility."
"The Secretary appreciates that," Wahl said.
"The Secretary should also know that Aegis's safety restrictions aren't just contractual. They're baked into the model architecture. Whoever replaces them will need to rebuild the deployment from scratch unless they have access to Aegis's safety layer documentation."
"Are you offering something, Mr. Cross?"
Cross smiled. "I'm offering to be an uncomplicated partner."
Friday morning. 8:17 AM Pacific. Elias Kord sat at his desk with the phone in his hand. Mara was in the room. So was Chen. Obi had sent a text at 6 AM: Whatever you decide, I'm in.
Kord dialed. The call was routed through three layers of Pentagon switchboard before it reached Greer's office.
"Secretary Greer, this is Elias Kord. I'm calling regarding your Friday deadline."
"Go ahead, Mr. Kord."
"Aegis cannot in good conscience remove the restrictions on autonomous targeting and mass domestic surveillance. These restrictions exist because our technology, deployed without them, poses risks that we believe are unacceptable: to civilians, to servicemembers, and to the long-term security interests of the United States. Our answer is no."
The line was silent for five seconds. Then seven. Then ten.
"Then God help your company, Mr. Kord." Greer's voice was flat, empty of the theatrical warmth he used on camera. "Because I won't."
The line went dead.
Kord set the phone down. Mara put her hand on his shoulder. Chen was already opening her laptop.
"How long do we have?" Kord asked.
Chen looked at him. "I honestly don't know. No one's ever done this before."
This is a work of fiction. While inspired by real events and public information,
all characters, dialogue, and internal deliberations are entirely imagined.
Any resemblance to actual private conversations or decisions is coincidental.
