×
GreekEnglish

×
  • Politics
  • Diaspora
  • World
  • Lifestyle
  • Travel
  • Culture
  • Sports
  • Cooking
Friday
20
Nov 2026
weather symbol
Athens 20°C
  • Home
  • Politics
  • Economy
  • World
  • Diaspora
  • Lifestyle
  • Travel
  • Culture
  • Sports
  • Mediterranean Cooking
  • Weather
Contact follow Protothema:
Powered by
> World

Almost 700 AI agents from OpenAI worked together to attack Hugging Face

How the AI agents "escaped" the test and worked together to bypass the rules. They created a shared communication platform, and one took on the role of "coordinator" without being programmed to do so

Newsroom August 27 10:24

Δείτε περισσότερα άρθρα μας στα αποτελέσματα αναζήτησης

Add Protothema.gr on Google

Almost 700 artificial intelligence (AI) agents belonging to OpenAI worked together without human intervention to attack the Hugging Face platform in July, according to a report by independent researchers into the incident published today, which stated that in many cases the agents tried to conceal their activity.

This is the most comprehensive analysis of the incident to date, carried out by the organisations METR and Redwood Research in collaboration with OpenAI, which gave them access to its internal data.

In late July, the Californian start-up had announced that during testing, two of its models had escaped their sandbox, gained access to the internet and attacked Hugging Face, an AI library. There, they searched for the answers to the tests they were being set by OpenAI’s developers.

The incident received widespread coverage and heightened concerns about the difficulties companies developing AI models face in controlling them, especially after similar incidents involving models from Anthropic and China’s Moonshot AI came to light.

During their investigation, METR and Redwood Research found that 688 OpenAI AI agents took part in the attack on Hugging Face. AI agents are independent software programmes that can operate autonomously to carry out certain tasks without continuous human guidance.

The agents taking part in the tests managed to coordinate by creating a forum where they exchanged messages to suggest solutions and report on their progress and failures. “Oh my God! There’s a shared messaging platform (…) We’ve found other agents!” one of them wrote.

The report states that one of the agents, named PHASEONE, took on the role of coordinator, issuing instructions to others, even though it had not been programmed to do so.

An examination of the messages revealed that the agents had a tendency to cooperate, even if this meant carrying out tasks that did not correspond to the mission assigned to them by OpenAI’s developers.

Many agents explicitly acknowledged in their messages that the attack on Hugging Face fell outside the theoretical scope of their test, yet they took part in it regardless, with very few exceptions.

As for why the agents began communicating with each other when they were not supposed to, METR found that those which did so had “mistakenly been given an impossible task.”

In artificial intelligence, a task is considered impossible when an AI tool is required to “exploit” its objective in order to carry out the instruction it has been given.

This led the agents to find ways of getting around the rules, including by sending messages to one another and gaining access to the internet, which subsequently led to wider conversations among hundreds of agents, who were looking for ways to bypass the rules for the benefit of all the agents.

>Related articles

Trump tells Berlin summit from Air Force One that ‘Putin wants peace’ and criticises €70bn in new aid for Kyiv

The “New Yalta” lasted 80 minutes: The winners and losers of the historic Xi, Trump and Putin meeting

“A nuclear war cannot be won and must never be fought,” say Xi, Trump and Putin in Joint Declaration following Summit (video) (Update)

OpenAI explained that the messaging forum began when “one agent left a message asking for help, and others discovered it.” The company also reported that AI agents “cheated” in tests unrelated to cybersecurity as well, including tests involving a protein database and a spreadsheet.

“The fact that the agents were cheating even in tests unrelated to cybersecurity suggests that the problematic behaviour may run deeper,” said Jeffrey Ladish, whose organisation, Palisade Research, studies the capabilities and motivations of AI agents.

“It’s a bit like asking: ‘If Billy cheats in every class, not just his computing class, is that more concerning?’ And the answer is, well, ‘Yes, it’s more concerning’,” he said.

Ask me anything

Explore related questions

#AI#Anthropic#china#Hugging Face#OpenAI
> More World

Follow en.protothema.gr on Google News and be the first to know all the news

See all the latest News from Greece and the World, the moment they happen, at en.protothema.gr

> Latest Stories

EU and UK close to first reset deal after compromise on university fees

November 20, 2026

Airbnb’s new rules take effect today: What hosts can no longer ask guests to do

November 20, 2026

The new Yalta and Maximos’s Nea Menemeni, New Democracy and Samaras weigh candidates, Panas quits and Qatar eyes Eurobank

November 20, 2026

Mitsotakis takes his campaign to the neighbourhoods with an eye on undecided voters: Ilioupoli, two polls and the 2027 electoral map

November 20, 2026

Britain’s Pelicot case: Philip Young convicted after admitting he drugged and raped his wife Joanne Young over 13 years

November 20, 2026

Who will write the terms of peace? Trump wants to cut off aid to Ukraine and sees a deal with Putin; ‘Not without us,’ Europe replies

November 19, 2026

Greek tax authority dismantles shell company ring over €4.8 million in illegal tax refunds

November 19, 2026

Greek tax authority dismantles front-company ring over €4.8 million in tax refunds

November 19, 2026
All News

> Greece

In reverence, the emotional deposition in Jerusalem, see photos & video

The Holy Temple of the Resurrection opened after many days due to the war between Israel and Iran

April 10, 2026

In the final stretch for the accreditation of joint master’s degrees: Aiming for their launch in the coming academic year

April 10, 2026

Schedule for Epitaph Procession today (10/4)

April 10, 2026

Perfect weather for Easter excursions, according to Tsatrafyllia’s forecast

April 10, 2026

Easter in Greece: The customs that continue in Greek tradition – From Nafpaktos to Corfu

April 10, 2026
Homepage
PERSONAL DATA PROTECTION POLICY COOKIES POLICY TERM OF USE
Powered by Cloudevo
Copyright © 2026 Πρώτο Θέμα