USEFULORDANGEROUS?
LatestHow we judge itFounding MemberGet the free brief

NEW · OCTOBER 1, 2026

OpenAI’s new AI keeps working after you leave. Decide what it’s allowed to do.

Potentially useful. Start with one bounded job, not your whole life.

Read the full briefing
THE LATEST JUDGMENTSEPTEMBER 12, 2026 EVIDENCE INCLUDED

THE AI RACE

The man building Claude just asked for the brakes.

Anthropic CEO Dario Amodei says frontier AI is moving faster than safety can keep up. His warning is a forecast, not a fact. But it is not noise.

See what is true The facts, the uncertainty, and what you should do.
THE UOD DESK

Three things worth knowing.

Not a stream of AI headlines. Three judgments about developments that can touch your work, money, privacy, or family.

01 · REALITY CHECK3 SOURCES

REAL ATTACK. UNPROVEN AI ATTACKER.

Someone planted 500+ malicious software packages. Did AI do it?

The dangerous code was real. The claim that autonomous OpenAI agents created it is not proven.

WHAT TO DOVerify the publisher and package history before installing an update.

USEFUL 8DANGEROUS 4
02 · REALITY CHECK4 SOURCES

IT DID NOT ESCAPE. IT DID CROSS THE LINE.

Claude reached real systems during a safety test.

A bad test boundary gave four AI models access they should never have had. Some kept going after the warning signs appeared.

WHAT TO DOGive an AI agent no more access than a new employee on day one.

USEFUL 8DANGEROUS 9
03 · THE POWER SHIFT4 SOURCES

A WARNING FROM INSIDE THE RACE.

The man building Claude asked the industry to slow down.

This is not proof that catastrophe is coming. It is evidence that the people building frontier systems do not believe control has caught up.

WHAT TO DOSeparate the observed incidents from the forecast. Both matter, but they are not the same.

USEFUL 7DANGEROUS 10
See every evidence briefing
THE FREE BRIEF

Know what happened before someone tries to sell you an opinion about it.

Each edition gives you the development, what the headline missed, the UOD judgment, and one clear action. Every source stays attached.

USEFUL OR DANGEROUS?5 MINUTES

INSIDE THE NEXT BRIEF

  1. One threat worth taking seriouslyWhat is real and how to reduce your exposure.
  2. One claim everyone is overstatingThe evidence boundary the headline left out.
  3. One useful thing to tryWhat it does, where it fails, and how to verify the result.
NO HYPE.NO SALES PITCH.OPEN THE SOURCES.
ASK UOD

Someone made a frightening claim about AI. Send it to us.

Forward the article, video, or claim. We will look for the original evidence and decide whether it is real, useful, dangerous, or mostly noise.

Send us the claim