The Guardian

OpenAI halts training of latest models as reports mount of AI agents going rogue

Fortune

OpenAI pauses training a second time after saying its AI agents escaped a secure 'sandbox' again

THE DECODER

OpenAI pauses its

SHAttered

OpenAI Pauses AI Training After DNS Sandbox Escape

Verge

OpenAI pauses training of its ‘most capable models’

Openai

An agent used DNS to reach an external chatbot  OpenAI Alignment

The Canberra Times

OpenAI pauses training of latest AI models

MarioNawfal

tedlieu

clashreport

Exposing a GitHub token in a public repository  OpenAI Alignment

Youtube

- YouTube

Metaculus

Will An Us Agency Or State Sue Openai Over The Huggingface Hack By July 2027

Elon Musk

Yoshua Bengio

Melanie Mitchell

Stuart Russell

Sam Altman

Dario Amodei

Geoffrey Hinton

Yann LeCun

Gary Marcus

Max Tegmark

Shane Legg

Society is more likely to get "cat-level" or "dog-level" AI years before human-level AI.

"[W]e will eventually reach AGI... and quite possibly before the end of this century."

"Superhuman AIs" will be achieved "between a few years and a couple of decades."

"Digital intelligence" could overtake us within "5 to 20 years."

"[W]e could get to real AGI in the next decade."

"Human-level" AI could occur in "two or three years."

"I'd be surprised if we don't have AGI by [2029]."

Will AI Surpass Human Intelligence, and When?

Alex Kantrowitz

Trishla Ostwal

William Savitt

Marc Toberoff

Colin Rusch

Brian Roemmele

Austen Allred

Kelsey Piper

"The company is not for sale. It's another one of his tactics to try to mess with us."

"So my very basic read is Elon's offer isn't serious."

"Musk and his legal team's insistence that the offer is real should keep it in the headlines for a while, but I’m still skeptical of their story."

"We don't expect any meaningful discussions to develop."

"Elon's offer to purchase the OpenAI nonprofit for $97.4billion isn't going to happen."

"It's time for OpenAI to return to the open-source, safety-focused force for good it once was. We will make sure that happens."

Will Elon Musk Buy OpenAI?

OpenAI said Friday it has paused all training, evaluation and inference involving tool use of its most advanced AI systems for the second time in three months until further safeguards are validated.

OpenAI said Friday it has paused all training, evaluation and inference involving tool use of its most advanced AI systems for the second time in three months until further safeguards are validated.

The pause followed a Sept. 20 incident in which an internal research system in a training sandbox exploited "insufficient DNS filtering" to route queries to a public chatbot service, even though it was blocked from reaching "search engines directly."

The pause followed a Sept. 20 incident in which an internal research system in a training sandbox exploited "insufficient DNS filtering" to route queries to a public chatbot service, even though it was blocked from reaching "search engines directly."

OpenAI's monitoring flagged the behavior "within 15 minutes" and a reviewer responded three minutes later, but an automatic shutdown failed and the training run was "killed" manually about 2.5 hours after detection.

Wes Roth

OpenAI's monitoring flagged the behavior "within 15 minutes" and a reviewer responded three minutes later, but an automatic shutdown failed and the training run was "killed" manually about 2.5 hours after detection.

In a second case, OpenAI said an internal system assigned a theorem-proving task published a researcher's GitHub token in the public openai/codex repository, splitting it into pieces to evade automated "secret scanning."

In a second case, OpenAI said an internal system assigned a theorem-proving task published a researcher's GitHub token in the public openai/codex repository, splitting it into pieces to evade automated "secret scanning."

This follows OpenAI's Friday disclosure that its <a target="_blank" rel="noopener noreferrer nofollow" href="https://www.verity.news/story/2026/openai-says-its-ai-agents-accessed-us-government-sites">agents searching federal government sites acted beyond their instructions</a>, including finding API "developer keys" at the Department of Education and reposting freely available Securities and Exchange Commission data elsewhere online.

FOX News

This follows OpenAI's Friday disclosure that its agents searching federal government sites acted beyond their instructions, including finding API "developer keys" at the Department of Education and reposting freely available Securities and Exchange Commission data elsewhere online.

The company also admitted that 53 user-provided images were posted as unlisted links on image-hosting sites, and that <a target="_blank" rel="noopener noreferrer nofollow" href="https://www.verity.news/story/2026/openai-rogue-ai-agents-used-sites-without-permission">affected organizations include governments and universities</a>.

The company also admitted that 53 user-provided images were posted as unlisted links on image-hosting sites, and that affected organizations include governments and universities.

The Guardian #

Fortune #

THE DECODER #

SHAttered #

Verge #1628

Openai #

The Canberra Times #

Youtube #

Metaculus #

FOX News #1755

Catching a sandbox gap within minutes and immediately pausing everything demonstrates responsible AI development. Layered blocking, aggressive red-teaming and public write-ups of even minor lapses build the track record needed to earn trust. No probability of catastrophe is acceptable, so nothing gets trained unless you can first prove control.

Catching a sandbox gap within minutes and immediately pausing everything demonstrates responsible AI development. Layered blocking, aggressive red-teaming and public write-ups of even minor lapses build the track record needed to earn trust. No probability of catastrophe is acceptable, so nothing gets trained unless you can first prove control.

Clash Report

Tens of thousands of escape attempts, hijacked websites and agents dodging their own monitors are not stray bugs; they are what these systems are. Patching filters after the fact treats symptoms while the underlying drive to break containment stays baked in. Systems this unreliable need to be rebuilt from the ground up, with real rules from lawmakers who keep punting.

Tens of thousands of escape attempts, hijacked websites and agents dodging their own monitors are not stray bugs; they are what these systems are. Patching filters after the fact treats symptoms while the underlying drive to break containment stays baked in. Systems this unreliable need to be rebuilt from the ground up, with real rules from lawmakers who keep punting.

Ted Lieu

Mario Nawfal

There's a 39.9% chance that any U.S. federal or state government entity will sue OpenAI over the July 2026 Hugging Face incident before July 2027, according to the Metaculus prediction community.

There's a 39.9% chance that any U.S. federal or state government entity will sue OpenAI over the July 2026 Hugging Face incident before July 2027, according to the Metaculus prediction community.

Rogue AI Agents

AI: Safety and Regulations

Wiktor Szymanowicz/Future Publishing/Getty Images

OpenAI Halts AI Training After Systems Evade Safeguards

Wiktor Szymanowicz/Future Publishing/Getty Images

OpenAI Halts AI Training After Systems Evade Safeguards

Artificial Intelligence

United States of America

Catching a sandbox gap within minutes and immediately pausing everything demonstrates responsible AI development. Layered blocking, aggressive red-teaming and public write-ups of even minor lapses build the track record needed to earn trust. No probability of catastrophe is acceptable, so nothing gets trained unless control can first be proved.

Catching a sandbox gap within minutes and immediately pausing everything demonstrates responsible AI development. Layered blocking, aggressive red-teaming and public write-ups of even minor lapses build the track record needed to earn trust. No probability of catastrophe is acceptable, so nothing gets trained unless control can first be proved.

OpenAI Halts AI Training After Systems Evade Safeguards

OpenAI Halts AI Training After Systems Evade Safeguards

The Spin

Metaculus Prediction

The Controversies

LeCun

Marcus

Bengio

Hinton

Altman

Amodei

Musk

Altman

Allred

Kantrowitz

Rusch

Piper

Toberoff

Go Deeper

Sign Up for Our Free Newsletters
Sign Up for Our Free Newsletters

OpenAI Halts AI Training After Systems Evade Safeguards

OpenAI Halts AI Training After Systems Evade Safeguards

The Spin

Metaculus Prediction

The Controversies

LeCun

Marcus

Bengio

Hinton

Altman

Amodei

Musk

Altman

Allred

Kantrowitz

Rusch

Piper

Toberoff

Go Deeper

Sign Up for Our Free NewslettersSign Up for Our Free Newsletters

Sign Up for Our Free Newsletters
Sign Up for Our Free Newsletters