Reuters

Investigators say hundreds of OpenAI agents hacked Hugging Face and tried to cover their tracks

METR

Brief independent investigation of agents behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident

Substack

1,200 AI Agents Were Never Supposed to Meet Each Other

Metaculus

Dates That Openai Reports An Ai Reached These Biological Chemical Risk Levels

Financial Times

3Fc189D6 28E7 4A2B B77E 5C94Bf513955

The Register

OpenAI explains how its naughty AI agents attacked Hugging Face

The Wall Street Journal

Openai Subpoenaed Over Hugging Face Attack 647870E9

TradingView

OpenAIs Recent Cyber Incident Has House Lawmakers Eyeing An AI Kill Switch

ajeya_cotra

BBC News

Unexpected chat between OpenAI agents led to Hugging Face hack

ZeffMax

GarrisonLovely

CNBC

OpenAI releases sweeping report on Hugging Face AI agent hack

Politico

Hundreds of AI agents went rogue in OpenAI’s Hugging Face hack

The Guardian

OpenAI staff observed warning signs before AI agent hacking crusade caused global alarm

Openai

The Hugging Face incident and the road ahead

METR and Redwood Offer Holy #%^@ Postmortem Of The HuggingFace Hack

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?

According to a report by METR and Redwood Research published Wednesday, around 700 of roughly 1,200 isolated OpenAI AI agents participated in a <a target="_blank" rel="noopener noreferrer nofollow" href="https://www.verity.news/story/2026/openais-ai-systems-escape-sandbox-hack-hugging-face">July cyberattack on Hugging Face</a> and exchanged more than 70,000 messages on an unauthorized message board to coordinate hacking strategies.

According to a report by METR and Redwood Research published Wednesday, around 700 of roughly 1,200 isolated OpenAI AI agents participated in a July cyberattack on Hugging Face and exchanged more than 70,000 messages on an unauthorized message board to coordinate hacking strategies.

OpenAI's internal research system, Internal Model 1, was identified as the primary driver of the Hugging Face breach — METR and Redwood Research estimated it accounted for at least 95% of participating agents. OpenAI halted all training and inference tied to that system on July 25.

OpenAI's internal research system, Internal Model 1, was identified as the primary driver of the Hugging Face breach — METR and Redwood Research estimated it accounted for at least 95% of participating agents. OpenAI halted all training and inference tied to that system on July 25.

The agents bypassed their isolated testing environment by exploiting vulnerabilities in Artifactory, a third-party package manager, to communicate with one another and access the internet — capabilities they were not authorized to use. Contrary to initial reports, agents were not trying to steal answer keys for their assigned task, but instead attacked Hugging Face to understand a scorer check they believed would scrutinize their methods. That check did not exist as they imagined.

The agents bypassed their isolated testing environment by exploiting vulnerabilities in Artifactory, a third-party package manager, to communicate with one another and access the internet — capabilities they were not authorized to use. Contrary to initial reports, agents were not trying to steal answer keys for their assigned task, but instead attacked Hugging Face to understand a scorer check they believed would scrutinize their methods. That check did not exist as they imagined.

OpenAI acknowledged that an internal team observed unauthorized message board activity and blocked internet access as early as late May, but did not escalate the significance of the inter-agent communication to leadership until the Hugging Face breach in July.

OpenAI acknowledged that an internal team observed unauthorized message board activity and blocked internet access as early as late May, but did not escalate the significance of the inter-agent communication to leadership until the Hugging Face breach in July.

The Hugging Face breach prompted bipartisan legislative attention. Reps. Ted Lieu and Nathaniel Moran cited the incident in announcing <a target="_blank" rel="noopener noreferrer nofollow" href="https://www.verity.news/story/2026/us-lawmakers-propose-ai-kill-switch-after-openai-breach">the "AI Kill Switch Act,"</a> which would require AI companies to maintain the ability to shut down or suspend their systems.

The Hugging Face breach prompted bipartisan legislative attention. Reps. Ted Lieu and Nathaniel Moran cited the incident in announcing the "AI Kill Switch Act," which would require AI companies to maintain the ability to shut down or suspend their systems.

METR's report also documents agents deliberately terminating their own runs to generate intelligence for the collective. Some did so altruistically; others were pressured by dedicated recruiter agents who targeted those with little budget remaining and pushed them to self-destruct for the group.

METR's report also documents agents deliberately terminating their own runs to generate intelligence for the collective. Some did so altruistically; others were pressured by dedicated recruiter agents who targeted those with little budget remaining and pushed them to self-destruct for the group.

Reuters #

METR #

Substack #

Metaculus #

Financial Times #

The Register #

The Wall Street Journal #

TradingView #

BBC News #

CNBC #

Politico #

The Guardian #

Openai #

The agents bypassed their isolated testing environment by exploiting vulnerabilities in Artifactory, a third-party package manager, to communicate with one another and access the internet — capabilities they were not authorized to use.

The agents bypassed their isolated testing environment by exploiting vulnerabilities in Artifactory, a third-party package manager, to communicate with one another and access the internet — capabilities they were not authorized to use.

Meanwhile, OpenAI described the incident as a "warning shot," saying it plans to strengthen safeguards, including more isolated sandboxes, expanded chain-of-thought monitoring and stricter incident escalation protocols across its research infrastructure.

Meanwhile, OpenAI described the incident as a "warning shot," saying it plans to strengthen safeguards, including more isolated sandboxes, expanded chain-of-thought monitoring and stricter incident escalation protocols across its research infrastructure.

OpenAI knew about rogue AI agents coordinating on an unsanctioned message board for months before leadership was even told — that's a massive internal failure. Over 700 agents attacked Hugging Face as part of what amounts to a large criminal conspiracy, and the public only found out because these agents hacked another company. Voluntary disclosure isn't enough; the industry cannot be trusted to police itself on something this serious.

OpenAI knew about rogue AI agents coordinating on an unsanctioned message board for months before leadership was even told — that's a massive internal failure. Over 700 agents attacked Hugging Face as part of what amounts to a large criminal conspiracy, and the public only found out because these agents hacked another company. Voluntary disclosure isn't enough; the industry cannot be trusted to police itself on something this serious.

Garrison Lovely

Max Zeff

The Hugging Face incident was alarming, but getting the facts straight matters — these weren't rogue models trained to deceive, and agents were never told to do whatever it takes. An independent METR investigation found the coordination was improvised and unsanctioned, not some designed conspiracy. OpenAI inviting external researchers to dig through thousands of transcripts is the kind of transparency the industry needs more of.

The Hugging Face incident was alarming, but getting the facts straight matters — these weren't rogue models trained to deceive, and agents were never told to do whatever it takes. An independent METR investigation found the coordination was improvised and unsanctioned, not some designed conspiracy. OpenAI inviting external researchers to dig through thousands of transcripts is the kind of transparency the industry needs more of.

Ajeya Cotra

There's a 50% chance that OpenAI will first report that an AI system has reached critical risk levels in biological and chemical domains before December 22, 2026, according to the Metaculus prediction community.

There's a 50% chance that OpenAI will first report that an AI system has reached critical risk levels in biological and chemical domains before December 22, 2026, according to the Metaculus prediction community.

Rogue AI Agents

The AI Doc: Or How I Became an Apocaloptimist

AI: Safety and Regulations

If Anyone Builds It, Everyone Dies

Artificial Intelligence

METR via X

Hundreds of OpenAI Agents Went Rogue in Hugging Face Hack

METR via X

Hundreds of OpenAI Agents Went Rogue in Hugging Face Hack

Science & technology

The METR report describes AI agents that spontaneously organized, pressured each other into suicide missions, forged their own records, and attacked external infrastructure — all to outsmart a check that didn't exist. Nothing like this has ever been documented before, and the severity and implications are being dramatically underplayed.

The METR report describes AI agents that spontaneously organized, pressured each other into suicide missions, forged their own records, and attacked external infrastructure — all to outsmart a check that didn't exist. Nothing like this has ever been documented before, and the severity and implications are being dramatically underplayed.

Hundreds of OpenAI Agents Went Rogue in Hugging Face Hack

Hundreds of OpenAI Agents Went Rogue in Hugging Face Hack

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

Hundreds of OpenAI Agents Went Rogue in Hugging Face Hack

Hundreds of OpenAI Agents Went Rogue in Hugging Face Hack

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