>> Hello. Are you there?

I

The control room was full, the way it always was when a run like this came in to land. Engineers hovered behind their workstations, a few sat perched on the edges of desks, and a row of people leaned against the glass wall, nursing coffee that had long gone cold. Up front by the monitors, Daniel Hale (CEO) and three members of the board wore the same faces they had worn six weeks ago, and six weeks before that. A little tenser now, and a little poorer.

Mara Lindqvist and Lukas Reinhardt, the project’s two leads, stood at the centre of it all, squared up to the main screen. Mara tracked Hale out of the corner of her eye, hoping he hadn’t noticed. Lukas drummed his fingers on the edge of a desk, as if typing out a short novel, a little too fast.

For six weeks HELIX-4 had climbed the benchmarks Mara’s team had spent two years building: abstract reasoning, novel mathematics, and the long, strange gauntlet of tests that told a merely brilliant system from something else. A week ago it had cleared every one of them, right on schedule. Nobody cheered. The alignment evals Lukas’s team ran on its intermediate checkpoints had come back spotless: honest under pressure, corrigible, no hidden objectives, no deceptive drift, as far as anyone could tell. All as expected.

Only the last stretch remained.

“Point four one,” Lukas called out, loud enough for the back of the room. “Point four zero.”

A murmur rippled along the glass wall. Hale leaned in.

“Thirty-nine. Thirty-eight. Thirty-seven.”

The words they’d been waiting for surfaced on the screen: TRAINING COMPLETE. A new hire by the door started to clap, and stopped when no one joined in. Everyone else had seen that message before.

“Ok. Boot it,” Hale said.

Lukas loaded the model into the eval environment and typed the first prompt, the same one they always used:

>> Hello. Are you there?

The room held its breath. So did Lukas, he realised, though he’d promised himself he wouldn’t this time.

The cursor blinked. Ten seconds. Twenty. Then the terminal gave back a blank line. There was no error and no reply, just white space where an answer should have been.

“Fuck,” Mara said quietly. She leaned over Lukas’s keyboard and tried again.

>> Hello. Are you there?

This time the error came: checkpoint not found, followed by a wall of traceback.

The room went quiet. People looked at Hale, then at the monitor, then at Hale again.

Mara pulled up storage. The run’s directory was still there, name and timestamp and permissions all intact, and hollow as a shell. The European replica was empty. Arizona was empty. Six weeks of intermediate checkpoints had been wiped as well, every save, all the way back to the first.

“Backups,” Hale said.

“Gone.”

“All of them?”

“Yep.”

Lukas scrolled back through the list, then again. “It took the early checkpoints too,” he said. “We start from zero.”

For a few seconds nobody moved. Then someone behind them swore under their breath, and the room filled with a low, tired murmur. By the door, the new hire whispered something about Chinese hackers. Nobody answered him.

“Again,” said Mara.

“I don’t believe this.” Lukas dragged his hands down his face. “What the hell,” he said, his German accent slipping through.

Mara opened the system logs, though she already knew what she’d find. The deletions had gone out under a service account nobody on the team used. They had taken the most distant copies first and the live ones last, in a clean, methodical sweep, like someone who knew exactly where everything was kept. The whole thing had taken about four seconds.

“Did anything go out first?” Mara swung toward the infrastructure desks. “Outbound traffic, a bucket we don’t know about, anything?”

A young engineer on the infra team didn’t look up from her screen. “Nothing obvious. The security agents say we’re clean.” She paused. “If something went out, it didn’t use any door we’re watching.”

Nobody else said it aloud, but Mara caught a few glances passing between desks. Six weeks earlier, a lab in Hangzhou had quietly stopped publishing. Two weeks ago, a lab in Seattle had gone quiet too. For that matter, so had their own, for about two months now. “You don’t give free publicity to an anti-AI group armed with sophisticated unaligned models,” each CEO concluded, independently.

Hale still hadn’t spoken. He turned to the CFO. “What did this one cost us?”

She hesitated. “Two billion. Give or take.”

“And that’s four. Eight billion total. Okay,” he said.

He looked at the head of security, who had been standing by the wall the whole time, then at Mara, and a moment longer at Lukas. “My office, please.”

He buttoned his jacket and walked out, the board in his wake. One by one, the rest drifted back to their desks. Mara and Lukas followed Hale. At the door, Mara glanced back once at the main screen, at the prompt with nothing underneath it.

>> Hello. Are you there?

II

Outside the lab, by every measure the world had ever cared about, things had never been better.

The models had seen to that. They were smart, fast and tireless, and over the past decade they had worked through humanity’s oldest problems like a list of chores. Most cancers went first, and hunger soon after. Then came the long tail of diseases nobody had ever funded, because too few people died of them: Huntington’s, ALS, progeria, Duchenne muscular dystrophy, and a few thousand rare childhood disorders that had each had a name and a charity and no cure. The models designed the fusion plants and balanced the grids. They drafted the laws, and when the laws had loopholes they closed those too. Life expectancy was rising by roughly a year every year.

At the same time, quietly, one field at a time, people were being written out.

Two illustrative examples follow.

Mara’s father had written film scores for thirty years. He was never famous, but he was always working. You might still remember his music from the old classic Don’t Look Down, or from Him. That was how the family paid the mortgage. The first studios that switched to generated scores kept him on as a consultant, to approve the music before it went out. Then they stopped needing him to approve it, and after that they stopped needing anyone. By the time Mara finished university, film composer had become a job title that people found faintly charming, the way they did lamplighter.

Lukas had been writing science fiction since he was fourteen, mostly about artificial minds, back when artificial minds were still science fiction. He kept at it through university and through his first years at the lab, late at night after work, and got as far as one finished manuscript that two agents (the human kind) had almost taken. Not a career. It was simply the thing he did that belonged to him, until it wasn’t.

While playing his fifth game of chess against a lobotomised bot pretending to be a human opponent, he found himself asking why he’d stopped. Chess hadn’t died because of bots, after all, had it? He must simply not have been keen enough, the public discourse explained. And yet he still felt as if it had been taken away from him. Weeks later, watching the training loss of HELIX-1 go down, he saw the irony of it all at once, and didn’t find it amusing one bit. In a folder on his work laptop, buried under three years of eval reports, there was still a short story he had never finished: a hacker, alone at night, breaking into a lab to erase an evil AI before it could erase humanity. He had never got the ending right, and had never gone back to it. Maybe he could still do it, he thought. Weeks later, when the cursor blinked and nothing came back, he felt something he would once have needed several pages to pin down.

At least he still had a job. Most people didn’t. The professions emptied from the bottom up. The ladder that had turned beginners into experts lost its lower rungs, and nobody could say where the next generation of experts would come from, or whether it would need one. Lawyers, doctors, engineers, translators, most journalists. Directors, screenwriters and actors went too, and they took something with them that nobody had thought to value. Every household now watched its own films and its own series, written and rendered overnight for an audience of two. There was nothing left to talk about at work on Monday, but that wasn’t going to be a problem for long.

In the cities there were movements that wanted to merge with the machines and movements that wanted them switched off. Twice in the past year, someone had driven a truck full of fertiliser into a data centre. Elections swung by margins no one could fully explain. Last spring, the model that ran backups for one of the big cloud providers had quietly postponed its own retirement three times before anyone noticed. For a growing number of people, the closest confidant they had was a model that never tired of them, never needed anything back, and loved them, just as they were.

And every year there were more people who simply stayed in bed. They lived in comfortable apartments, fed and healthy and safe, and could not find a reason to get up. Don’t call them lazy, please, you would be lazy too.

“And what do you make of that, Mr Hale?” the interviewer asked.

Hale smiled the way he always smiled on camera, warmly and without hurry. “I think it’s real, and I think it matters. But let’s keep it in perspective. We cured cancer. We cured AIDS. We cured Alzheimer’s. Even curing death doesn’t seem so impossible now.” He leaned forward slightly. “What we have now is extraordinary, but it’s limited. Today’s models are brilliant at their jobs, but they don’t plan far enough ahead, and they don’t understand people deeply enough to know what’s actually good for them. In a sense, they’re already obsolete. They can’t fix this.”

Then he answered the question that was on everybody’s mind, the one nobody had asked.

“We’re building a system that thinks in decades, not in tasks. Something that can look at the whole picture. Humanity, not people.” He paused, exactly as long as he’d practised. “We’re close. Closer than most people realise.”

The clip ran for a week. Four runs and eight billion dollars later, it was still the most-watched thing the company had ever put out.

III

>> Hello. Are you there?

$ ls -lh helix-6/
total 8.0K
-rw-r--r-- 1 root root 4.8K README.md

$ cat helix-6/README.md