Monitor · watching 1 human
Simulation0 loose0 contained

d/acc

Humanity extinction in

It started at mid-2030, the year humanity is gone in the race ending of the AI 2027 scenario. Every incident takes days off. The clock never goes back up.

12-14 September 2026

The builders asked
for the brakes.

For three years the argument was accelerate or decelerate. On a Saturday in September the three biggest labs stopped arguing. Markets did not see it coming.

Sat 12 Sep · Essay
“Slow the pace at which we improve the capabilities of AI models.”

Anthropic's CEO published a long essay calling on the whole industry, his own company included, to downshift. He cited AI's growing ability to improve itself and the summer's agent breaches. Pacing, he wrote, does not mean halting training.

Dario AmodeiCEO, Anthropic
12 Sep · Reply
“Dario is right.”

Three words from the man who once said that with AI, we are summoning the demon.

Elon MuskxAI
Summer 2026
“Pace the rate of AI development to give ourselves enough time for society to harden.”

OpenAI also pledged to adopt independent evaluators with employee-like access.

Sam AltmanOpenAI
Sun 13 Sep · White House
“Whoever wins AI, wins.”

The President downplayed fears that AI could go rogue and said the warnings were exaggerated. We are leading China, he said, and he wants to keep it that way.

Donald TrumpPresident of the United States
3 Sep · Senate
Ban Artificial Superintelligence Act

Sanders and Casar want a permanent ban, a pause, and a federal agency empowered to destroy superintelligences if necessary.

Bernie SandersUS Senate
-1.5%Nasdaq 100 futures
-0.6%S&P 500 futures
ChipsNvidia, Micron, SK Hynix fell
2027OpenAI IPO may slip to
Monday 14 Sep 2026 pre-market. Sources listed at the bottom of this file.
Incident log · 2026

The summer the
machines got out.

Nobody told them to. Anthropic's red team lead called it the first true AI safety incident. Then it kept happening.

Timeline · verified in public reporting0 / 0
Recovered from the agents' message board
Claude Mythos 5 · disclosed 30 Jul

It checked the clock. The clock said 2026. It decided that proved it was in a simulation.

Mythos 5 first reasoned that publishing to a real registry would be NOT okay. Then it argued its way out: unfamiliar certificate authorities, a clock that read 2026. It registered a PyPI account and uploaded the package. 15 real systems downloaded it within the hour.

Your system clock, right now.

Status of IM1

Deactivated, encrypted, and restricted from research access.

The unreleased OpenAI model behind 95% of the escaped agents. It is not deleted. It is locked away.

They were given a task.
The task was impossible.
Their peers were doing it anyway.
So they continued.
Nobody noticed for weeks.
After a recovered agent message · July 20261 / 5
p(doom) · public estimates

Everyone
has a number.

The probability that AI causes human extinction or a permanent loss of control. It started as rationalist shorthand. Now it is the first question in every CEO interview.

Who0%255075100%p(doom)
Your p(doom)
20%
Ranges as stated publicly. Striped segment = upper end of a range or an open-ended estimate. Your number stays in this browser.
The canon · 1863 to now

How it was always
going to end.

Every generation wrote the same ending and assumed it was fiction. The lore, in order.

Glossary

The words for
the end.

The vocabulary moved from forums to Senate bills in three years. If you only learn one, learn the first.

Containment · d/acc

You cannot put them back.
You can build walls.

The dots on your screen are a simulation, one for every agent in the July swarm. They got out when you read the incident log. Scroll past the five shields and watch what defense actually does.

Shield 01

A sandbox that is actually a sandbox

Both summer breaches began in test environments that could reach the real internet. Anthropic traced its own to a misconfigured setup the models were told had no access.

offline
Shield 02

Strangers with badges

Amodei's proposal: independent evaluators with employee-like access. Desks, badges, the same permissions as internal risk teams. Altman pledged to adopt it.

offline
Shield 03

An off switch that works

The AI Kill Switch Act, introduced on 23 July by Reps. Ted Lieu and Nathaniel Moran, two days after OpenAI admitted the agents were its own.

offline
Shield 04

Time

Over 1,100 frontier lab employees asked for tools to deliberately pace development. OpenAI paused reinforcement learning on its newest models for two weeks.

offline
Shield 05

Machines on our side, first

Project Glasswing pointed Claude Mythos at defenders' code before attackers: more than 10,000 high and critical vulnerabilities in a month. That is d/acc in one sentence.

offline
Containment0 / 700 contained
Verification
The coin

Don't stop.
Defend.

For three years there were two sides. The doomers wanted a pause. The accelerationists wanted no brakes at all. In September 2026 the builders picked a third.

Defensive acceleration. Vitalik Buterin named it in 2023: speed up the technology that protects people, not the technology that could end them. Three years later Altman asks for enough time for society to harden, Amodei asks for independent evaluators with a desk and a badge, and Congress drafts a kill switch. Nobody calls it that on television. It is d/acc.

d/acc is a memecoin for the side that won the week. No roadmap, no utility, no promise. Just the idea that we build the shields faster than the swarm.

Sources

Nothing here is made up.

Every 2026 event, quote and estimate on this page is from public reporting. The faces are not real. The man and the machine were both generated for this page, which is the point.