The Index
The director of a prestigious AI research institute discovers that the annual report her team is about to publish contains a number that will reshape geopolitics — and three governments are trying to make sure it never sees daylight.
The Index
Dr. Helen Zhao had not slept in thirty-one hours, and the number on her screen had not changed.
2.7.
She stared at it the way a doctor stares at a lab result that confirms the diagnosis no one wanted. The number itself was not complicated. It was a percentage. It represented the performance gap between the top-ranked American AI model and the top-ranked Chinese AI model on the benchmark her institute had spent seven years establishing as the industry standard. Last year, that number had been 12.3. The year before, 18.5.
2.7 was not a gap. It was a rounding error.
Helen pushed back from her desk and looked out the window of her office on the fourth floor of the Cordura Hall annex. The campus was quiet at 2 AM. A few lights burned in the computer science building across the quad, where graduate students were probably training models of their own, blissfully unaware that the report sitting on Helen's laptop was about to make their work a matter of national security.
Her phone buzzed. She looked at it and felt something tighten in her chest.
The caller ID said RESTRICTED, which on her phone meant one of exactly four numbers.
"Dr. Zhao, this is Margaret Winslow at OSTP."
Office of Science and Technology Policy. The White House.
"It's two in the morning, Margaret."
"I'm aware. Have you finalized the index numbers?"
Helen closed her eyes. "The report publishes in eleven days."
"That's not what I asked."
"The numbers are final."
There was a pause that lasted long enough for Helen to hear someone else breathing on the line. Margaret was not alone.
"Dr. Zhao, we'd like to discuss the methodology before publication. There may be nuances in the benchmark design that warrant a supplementary analysis. We could send a team from NIST to assist."
Helen had run this institute for nine years. She had weathered pressure from tech companies who disliked their benchmark rankings, from foreign governments who disputed methodology, and from her own university administration who periodically suggested that perhaps the index could be "more constructive." She had never changed a number.
"The methodology is the same methodology we've used for four years. It's peer-reviewed. It's transparent. If NIST wants to publish their own analysis after our report, they're welcome to."
"Helen." Margaret's voice dropped the professional warmth. "The number you're about to publish will be on the front page of every newspaper in the world. It will be quoted in congressional hearings. It will appear in Chinese state media as evidence that their AI strategy is working. The geopolitical implications are —"
"Are not my responsibility. My responsibility is accuracy."
The other person on the line spoke for the first time. Male voice, older, the cadence of someone accustomed to being listened to. "Dr. Zhao, no one is asking you to change a number. We're asking you to contextualize it. A 2.7 percent gap means something different depending on whether you're measuring capability, deployment, economic impact, or strategic advantage. Your headline number doesn't capture those distinctions."
"Which is why the report is four hundred pages long."
"Nobody reads four hundred pages. They read the headline."
Helen looked at 2.7 on her screen. He was right about that part. Nobody would read four hundred pages.
The second call came at 6:15 AM, while Helen was making coffee in the institute's small kitchen. This one was from the Chinese consulate in San Francisco.
"Dr. Zhao, this is Vice Consul Liu Wei. I hope I am not disturbing you."
"What can I do for you, Vice Consul?"
"We have seen preliminary indications that your upcoming report will show significant progress in Chinese AI capabilities. We wanted to express our appreciation for your institute's rigorous and fair methodology."
Helen poured coffee and said nothing.
"We would also like to offer our cooperation for future reports. We can arrange access to additional Chinese research facilities, provide updated publication data, and facilitate interviews with senior researchers at our national laboratories. Full transparency."
"That's very generous."
"We believe in the importance of accurate measurement."
Helen understood what was being offered. Not cooperation. Leverage. If she accepted Chinese government facilitation for future reports, her institute's independence would be compromised in ways that were impossible to unwind. Every future number would carry an asterisk in the minds of anyone who knew about the arrangement.
"Vice Consul, our methodology relies on publicly available data and independently verifiable benchmarks. We don't accept facilitated access from any government, including our own."
The warmth in Liu Wei's voice cooled by exactly one degree. "Of course. We merely wanted to ensure you have the most complete picture possible."
The third call came at 11:30 AM. Helen was in a conference room with her deputy director, Raj Patel, reviewing the final layout of the report.
"Unknown number, DC area code," Raj said, glancing at her phone.
Helen picked up. "Zhao."
"Dr. Zhao, this is someone who would prefer not to be named. I'm calling on behalf of a bipartisan group of senators who will be holding hearings on AI competitiveness next month."
"I'm listening."
"Your report will be the central exhibit. The 2.7 number — we've seen the draft — is going to drive the narrative. Several senators have concerns that the number, while technically accurate, may be misleading without additional context."
"What context?"
"The investment differential. $285 billion versus $12 billion. If you led with the investment gap and presented the performance gap as a secondary finding, the narrative would be more balanced."
Helen looked at Raj. He was reading her expression and slowly shaking his head.
"The performance gap is the primary finding because performance is what the index measures. Investment is a supporting data point."
"Dr. Zhao, I want to be direct. If the report publishes with 2.7 as the headline, the political response will be severe. There will be calls for new export controls, new restrictions on academic collaboration, potentially new limits on foreign student enrollment in AI programs. Some of those measures would directly affect your institute's ability to operate."
"Are you threatening my funding?"
"I'm describing consequences. There's a difference."
Helen hung up.
Raj was quiet for a long time after she told him about all three calls.
"Three governments in twelve hours," he said finally. "That might be a record."
"The American government wants us to bury it. The Chinese government wants us to amplify it. And Congress wants us to reframe it so they can use it as ammunition for whatever they were already planning to do."
"And what do you want?"
Helen looked at the report on the screen. Four hundred and twelve pages. Seven years of data. Forty-three researchers across twelve institutions. The most comprehensive annual assessment of global AI capability ever produced.
And one number that had changed everything.
"I want to publish it exactly as it is and let the data speak."
Raj nodded. "Then we publish."
"Set the embargo for Sunday. No previews, no pre-briefings, no courtesy calls. The report goes live at midnight UTC and everyone sees it at the same time."
"The White House will be furious."
"The White House will survive. So will Beijing. So will Congress." She paused. "The number is 2.7, Raj. It was 18.5 three years ago. The trajectory is the story, and no amount of context or framing changes what that trajectory means."
"What does it mean?"
Helen looked out the window at the campus. Students were crossing the quad, backpacks slung over shoulders, phones in hand, living in a world shaped by decisions made in rooms they would never see.
"It means the race is real. And pretending otherwise — for any government, for any reason — is more dangerous than any number we could publish."
She turned back to the screen and began writing the executive summary.
The 2026 AI Index finds that the performance gap between the leading US and Chinese AI models has narrowed to 2.7 percentage points, down from 18.5 percentage points in 2023. US and Chinese models have traded the lead position multiple times since early 2025. The era of unchallenged American AI dominance is over.
She read it twice. Changed nothing. Saved the file.
The number had not changed. But by Sunday, everything else would.