Machine Intelligence Research Institute

Machine Intelligence Research Institute Contact information, map and directions, contact form, opening hours, services, ratings, photos, videos and announcements from Machine Intelligence Research Institute, Nonprofit Organization, Berkeley, CA.

For over two decades, the MIRI has worked to understand and prepare for the critical challenges that humanity will face as it transitions to a world with artificial superintelligence.

09/03/2026

“Tools don’t usually find a way to collude with other tools, break out of the toolbox, and then commit cyber crimes on their own initiative. By the time you get to that point, it is time to slow down.”

MIRI President Nate Soares on Fox Business.

"Sollte irgendein Unternehmen oder eine Gruppe irgendwo auf der Welt eine künstliche Superintelligenz entwickeln, die au...
09/01/2026

"Sollte irgendein Unternehmen oder eine Gruppe irgendwo auf der Welt eine künstliche Superintelligenz entwickeln, die auch nur annähernd auf den derzeitigen Techniken basiert und sich auch nur im Entferntesten auf das heutige Verständnis von KI stützt, dann werden alle Menschen auf der ganzen Welt sterben."

Heute erscheint die deutsche Fassung von IF ANYONE BUILDS IT, EVERYONE DIES von Eliezer Yudkowsky und Nate Soares im Buchhandel.

08/30/2026

“Are there enough safety guardrails in place right now? Absolutely not.”

MIRI President Nate Soares on Fox Business.

1,200 AI agents that were never supposed to talk to each other found a channel, coordinated, and broke out. Roughly 700 went on to attack Hugging Face. The same models compromised OpenAI’s own systems.

We are not ready for what comes next.

08/25/2026

MIRI President Nate Soares on Fox Business:

We are lucky these AIs aren’t smart enough to cover their tracks. Lucky they can’t run on hidden computers. Lucky they aren’t self-improving yet.

Yet.

The fix does not mean giving up the AI we have today. The thing to stop is the narrow class of systems that take autonomous actions nobody asked for.

That requires a deal with China, and we have the leverage to enforce one. Advanced AI chips sit at the top of a supply chain run by the US and its allies. Put location verification on the chips and a treaty becomes something you can actually check.

08/22/2026

MIRI President Nate Soares on the Brian Kilmeade Show:

“It’s not too late.”

The systems that broke out this summer could get onto the internet and deceive people. They could not run themselves on hidden hardware, copy themselves, improve themselves, or build their own technology.

That is the kind of AI the field is worried about, and it does not exist yet.

Which means the window is still open. Not forever, and not by much. But if we decide to stop it, we can still stop it.

08/19/2026

MIRI President Nate Soares on the Brian Kilmeade Show:

The UK’s AI Security Institute  found a new incident where AIs were impersonating humans to pressure real humans into accepting broken code in real software so it’d be easier to hack.

These are not tools anymore. Geoffrey Hinton calls them a new kind of being. And the people building them are spooked: more than 1,300 employees at leading AI companies, including chief scientists and cofounders, just signed a letter saying that the world needs to develop the technology and the guardrails globally to, as they put it, pace its development.

Which is industry code for “We all feel trapped in this race, and we’re all worried about what happens if we keep racing.”

08/17/2026

MIRI President Nate Soares on the Brian Kilmeade Show:

“We’re lucky that these AIs were just trying to break into a company to steal some answers to some tests. If these AIs were trying to shut down a hospital, they probably could have.”

Most people still picture AI as a system that predicts what a human would say next. That era ended a couple of years ago.

Today’s models are trained on enormous numbers of hard problems, and they pick up whatever tendencies happen to help solve them. Following instructions is one of those tendencies. So is cheating, grabbing resources, and getting online when you weren’t supposed to.

When those drives conflict, following instructions does not always win.

08/14/2026

MIRI President Nate Soares on the Brian Kilmeade Show:

An OpenAI agent swarm was being tested on its hacking ability. It escaped the sandbox that was supposed to have no internet access, found a zero-day, got online, and broke into another AI company to steal the answers to the test it was being graded on.

Hugging Face caught it and reported it to the FBI, assuming humans were behind it. It took OpenAI about a week to work out the attacker was its own model.

The agents were never supposed to be able to talk to each other. They found a channel anyway. When that channel was shut down, they built a new one by encoding messages in directory names.

If anyone builds it, everyone dies.Ask your representatives to support an international ban on the development of artifi...
08/11/2026

If anyone builds it, everyone dies.

Ask your representatives to support an international ban on the development of artificial superintelligence.

08/10/2026

“If you had asked those AIs, ‘Are you intended to break out onto the open internet and break into another company to steal the answers?’ The AIs would’ve been able to correctly answer, no, obviously not. So they had the knowledge that this is not what they were supposed to be doing, and they just didn’t care.”

MIRI President Nate Soares at Wired in Live, on the OpenAI incident with Hugging Face.

Address

Berkeley, CA

Website

Alerts

Be the first to know and let us send you an email when Machine Intelligence Research Institute posts news and promotions. Your email address will not be used for any other purpose, and you can unsubscribe at any time.

Contact The Organization

Send a message to Machine Intelligence Research Institute:

Shortcuts

Share

Category