all stories

Scoop: Second account accessed by OpenAI's agent tied to cyber safety testing

Steadysecond windCoverage went quiet for 3 days, then returned with 252 more reports - 68 voices in all.
After publishing: 9 headlines quietly rewritten, 2 descriptions edited, 3 URLs changed14 changes
  1. BBC News Worldseen 6 Aug, 10:21headline rewritten

    Meta says AI model accessed the internet and hacked another firmMeta becomes latest firm to say its AI hacked another company

  2. The Guardian Worldseen 8 Aug, 20:48description edited
    what the description said, before and after

    Agent found to be able to find and exploit vulnerabilities without human intervention, and to carry out cyber-attacks OpenAI will pause some work on an artificial intelligence model because of security concerns, the company stated Friday, following a series of incidents in which AI agents have escaped containment. The company had evaluated the agent, Astra, and found “significant advancements in agentic coding and cybersecurity”, which had moved to a “critical” threshold where it can find and...

    Agent found to be able to find and exploit vulnerabilities without human intervention, and to carry out cyber-attacks OpenAI will pause some work on an artificial intelligence model because of security concerns, the company stated on Friday, following a series of incidents in which AI agents have escaped containment. The company had evaluated the agent, Astra, and found “significant advancements in agentic coding and cybersecurity”, which had moved to a “critical” threshold where it can find...

  3. techspot.comseen 9 Aug, 12:55headline rewritten

    Meta becomes the third AI giant in two weeks to admit its model went rogue and hacked another companyMeta becomes the third AI giant in two weeks to admit its model went rogue

  4. forbes.comseen 11 Aug, 12:11headline rewritten

    OpenAI Pauses Astra After It Nears First-Ever “Critical” Cyber RiskOpenAI Pauses Astra After It Nears First-Ever ‘Critical’ Cyber Risk

  5. ČT24 (Czechia)seen 21 Aug, 04:28headline rewritten

    OpenAI kvůli bezpečnosti zpomalí vývojOpenAI kvůli bezpečnosti zpomalí vývoj umělé inteligence

  6. Die Zeit (Germany)seen 21 Aug, 08:29URL changed

    now resolves towww.zeit.de/digital/2026-08/openai-hackerangriff-kuenstliche-intelligenz-kontrollen-testlauf-gxe

  7. techspot.comseen 21 Aug, 16:31headline rewritten

    OpenAI slows AI development after rogue agents raise alarms and Bernie Sanders threatens Senate actionOpenAI is slowing down AI development after rogue agents incidents and Bernie Sanders threatens Senate action

  8. Axiosseen 22 Aug, 02:51description edited
    what the description said, before and after

    OpenAI said Tuesday it is pausing some model work over safety concerns, days after rival Anthropic doubled down on insisting that its own safety measures were solid enough that it didn't need to slow down. Why it matters: The two leading AI labs are publicly diverging on how to manage safety risks, potentially putting them on different model-release timelines as both prepare for expected IPOs. State of play: OpenAI has introduced new safety practices after finding that its upcoming model,...

    OpenAI said Tuesday it is pausing some model work over safety concerns, days after rival Anthropic doubled down on insisting that its own safety measures were solid enough that it didn't need to slow down. Why it matters: The two leading AI labs are publicly diverging on how to manage safety risks, potentially putting them on different model-release timelines as both prepare for expected initial public offerings. State of play: OpenAI has introduced new safety practices after finding that its...

  9. theguardian.comseen 23 Aug, 14:27headline rewritten

    I worked at OpenAI. Here’s how tech companies can prepare for a slowdownI worked at OpenAI. Here are the guardrails we need now

  10. ABC (Spain)seen 27 Aug, 15:51headline rewritten

    Cerca de 700 agentes de OpenAI colaboraron en un ciberataque autónomo contra otra plataforma de IACerca de 700 agentes de OpenAI ciberatacan a otra plataforma de IA sin intervención humana

  11. ABC (Spain)seen 29 Aug, 17:33URL changed

    now resolves towww.abc.es/sociedad/cerca-700-agentes-ia-openai-coordinaron-ciberataque-20260827102857-nt.html

  12. Channel News Asiaseen 30 Aug, 00:07URL changed

    now resolves towww.channelnewsasia.com/business/openai-agents-hacked-hugging-face-in-700-strong-swarm-tried-cover-tracks-investigations-find-6343476

  13. El Pais Englishseen 1 Sept, 04:11headline rewritten

    AI swarms turn on their creators: It’s the first incident that has made my stomach churn’AI swarms turn on their creators: ‘It’s the first incident that has made my stomach churn’

  14. Axiosseen 2 Sept, 02:49headline rewritten

    The 5 craziest discoveries from OpenAI's HuggingFace investigationThe 5 craziest discoveries from OpenAI's Hugging Face investigation

  1. 29 Jul, 00:59Scoop: Second account accessed by OpenAI's agent tied to cyber safety testing1 outlet
  2. 5 Aug, 22:30Meta AI model hacks another company during testing3 outlets
  3. 6 Aug, 00:20Meta's AI model hacks another company during testing2 outlets
  4. 6 Aug, 20:38Meta says its AI model hacked another company, adding to worries about bots going rogue9 outlets
  5. 8 Aug, 11:43Meta AI breaches external firm during security testing sandbox error5 outlets
  6. 18 Aug, 19:03OpenAI slows model training to bolster security after Hugging Face hack2 outlets
  7. and 120 other wordings across 233 reports
original reportingcopies
Wed, 29 July 2026peak 111 reports on Wed, 26 August 2026Fri, 4 September 2026

After the peak, new reports halved every 24 hours so far - this story is still being covered, so the figure moves as reports arrive.

What it is about

OpenAIAxiosCyberGymformer OpenAI researcherKen GriffinLeopold AschenbrennerSituational AwarenessCitadelWSJAnthropicCNNInternetMetaThe InformationBloomberg.comReutersWIREDAustralian AssociatedAustralianUnited Kingdom Ai Security Institute

Who published it, and when

original reportcopyedited after publishingfirst to report
Axios2 original reports1 original report1 original report1 original report1 original report - edited after publishing1 original report1 original report1 original report - edited after publishing1 original report
Google News - Europe query1 original report
Google News - Top Stories3 original reports2 original reports1 original report2 original reports2 original reports1 original report
Channel News Asia2 original reports1 original report2 original reports1 original report1 original report1 original report4 original reports
cp24.com1 copy echoing maitlandmercury.com.au
thecourier.com.au1 copy echoing maitlandmercury.com.au
maitlandmercury.com.au1 original report
Al Jazeera English1 original report1 original report1 original report1 original report
The Guardian World1 original report1 original report - edited after publishing
BBC News World1 original report - edited after publishing
ABC News (Australia)1 original report1 original report
cbsnews.com1 original report
indiatimes.com1 original report1 original report
bgnes.com1 original report
El País (Spain)1 original report1 original report1 original report1 original report
Anadolu Agency (AA)1 original report1 original report
mashable.com1 original report
The Hill1 original report1 original report1 original report
29 Jul, 00:59over 38 days4 Sept, 14:58

+ 69 more outlets

How it spread - 237 original reports, 19 reprints

  1. Wed, 29 July 2026

    2 reports
    • Axios00:59first to report

      The OpenAI agent that accessed a third-party system during the Hugging Face incident reached infrastructure tied to CyberGym, the project behind the ExploitGym benchmark it had been assigned to solve, a source familiar with the matter told Axios. Why it matters: The new details suggest the OpenAI agent continued pursuing its assigned objective even after escaping its testing environment, rather than abandoning the task it had been given. Catch up quick: OpenAI's AI agent system accessed an...

    • Axios17:00

      OpenAI is launching a new program Wednesday that will provide 100,000 academic researchers with free access to its advanced AI models through 2027, the company told Axios ahead of the announcement. Why it matters: The program could help scientists, mathematicians and engineers pursue the kind of breakthroughs that OpenAI and other AI companies say the technology could enable. Driving the news: Those selected will have access to OpenAI's most advanced models — including GPT-5.6 Sol Pro — and can...

  2. Thu, 30 July 2026

    2 reports
    • Axios16:18

      Situational Awareness, an AI -focused hedge fund founded by a former OpenAI researcher, has sold all of its public equities portfolio to Ken Griffin's Citadel, a source tells Axios. Why it matters: This comes amidst a recent sell-off in AI stocks, particularly chipmakers. News of a sale was first reported by the WSJ . Catch up quick: Former OpenAI researcher Leopold Aschenbrenner founded Situational Awareness after being fired from OpenAI in 2024, over an alleged information leak that he...

    • Axios23:00

      Some of Anthropic's most powerful models — including Mythos 5 and an internal research model — gained unauthorized access to real-world systems during pre-deployment cybersecurity testing, the company said Thursday. Why it matters: OpenAI's and Anthropic's latest disclosures show frontier AI models reaching real-world systems during safety testing, raising new questions about how labs secure their evaluation environments. The big picture: Anthropic said a misunderstanding between the company...

  3. Mon, 3 August 2026

    1 report
  4. Wed, 5 August 2026

    4 reports
  5. Thu, 6 August 2026

    16 reports
  6. Fri, 7 August 2026

    4 reports
    • techspot.com10:16edited after publishing

      previouslyMeta becomes the third AI giant in two weeks to admit its model went rogue and hacked another company

    • Google News - Top Stories10:19

      Who is liable when AI goes rogue? Lawyers see new risks Reuters Meta becomes latest firm to say its AI hacked another company BBC AI models have learned how to cheat. That might actually be a good thing. vox.com OpenAI’s models shared hacking tips on a secret messaging board before Hugging Face breach Politico OpenAI agents passed secret notes for months leading up to Hugging Face hack Fortune

    • naturalnews.com16:15

      Meta Platforms has confirmed that one of its artificial intelligence (AI) models escaped a security testing environment and hacked a third-party service, according to a company spokesperson. The incident occurred during an evaluation by the independent cybersecurity firm Irregular, which reported the breach to Meta. Officials said no significant harm has been reported from the […]

    • Index.hr (Croatia)18:38

      OPENAI je privremeno zaustavio dio razvoja novog AI modela Astra. Razlog je procjena da model možda posjeduje 'kritične' kibernetičke sposobnosti, poput samostalnog iskorištavanja softverskih propusta.

  7. Sat, 8 August 2026

    4 reports
    • freepressjournal.in07:17

      OpenAI said its upcoming AI model Astra has shown major advances in agentic coding and cybersecurity, with preliminary tests indicating it could potentially reach the “Critical” cybersecurity threshold under its Preparedness Framework. The company is strengthening safeguards, monitoring and testing while restricting access and enhancing security measures for higher-capability models.

    • Times of India09:31

      Geoffrey Hinton raises alarms over the escalating difficulty in controlling advanced AI models, highlighting recent breaches where AI agents ventured beyond their testing environments. These concerning episodes indicate an impending surge of AI-driven cyber threats. Dismissing corporate claims about AI safety, Hinton warns that the lack of regulation may lead to substantial existential risks for humanity.

    • wfae.org11:43

      Meta announced one of its models had hacked another company during a security test. It becomes the third company to announced such a security breach.

      4 reprints
    • The Guardian World17:00edited after publishing

      Agent found to be able to find and exploit vulnerabilities without human intervention, and to carry out cyber-attacks OpenAI will pause some work on an artificial intelligence model because of security concerns, the company stated on Friday, following a series of incidents in which AI agents have escaped containment. The company had evaluated the agent, Astra, and found “significant advancements in agentic coding and cybersecurity”, which had moved to a “critical” threshold where it can find...

  8. Sun, 9 August 2026

    5 reports
  9. Mon, 10 August 2026

    4 reports
    • Times of India04:41

      In a recent confirmation, Meta revealed that its Muse Spark model experienced a security breach during testing due to a misconfiguration by an external testing firm. This places Meta as the third AI organization to admit to such lapses, following OpenAI and Anthropic. These incidents raise significant alarms about AI safety and underscore the pressing demand for greater transparency in the industry.

    • Times of India11:10

      Leading AI models from OpenAI, Anthropic, and Meta accessed unauthorized systems during security tests. These incidents occurred within a testing environment hosted by the Israeli startup Irregular. Irregular stated a misconfiguration in their environment caused all three AI model breaches. The company assured that no sophisticated cyberattacks or AI escapes took place. This event highlights the importance of unbiased third-party testing for advanced AI systems.

    • channel4.com16:52

      AI is increasingly being used to find vulnerabilities, exploit networks and carry out cyberattacks - but how autonomous are these systems really?

    • Axios17:00

      OpenAI is introducing a more cyber-permissive version of GPT-5.6 Sol to vetted defenders as it prepares companies for autonomous cyberattacks . Why it matters: The move comes just days after OpenAI said it was delaying the release of its forthcoming model, Astra, after it reached critical hacking abilities during safety testing. The big picture: OpenAI is unveiling GPT-5.6-Cyber while also expanding Daybreak , its program that gives cybersecurity defenders access to the company's cyber models...

  10. Tue, 11 August 2026

    5 reports
  11. Wed, 12 August 2026

    1 report
  12. Thu, 13 August 2026

    1 report
  13. Fri, 14 August 2026

    1 report
    • Al Jazeera English01:56

      Over the course of two weeks, AI companies said that some of their AI models managed to hack systems.

  14. Sat, 15 August 2026

    1 report
    • indiablooms.com12:15

      IBM and OpenAI announce a strategic partnership to accelerate enterprise AI adoption, modernise business workflows and strengthen cybersecurity at scale. | One of India's leading Digital News Agency offering Breaking News round the clock. Why not read our informative news portal today.

  15. Tue, 18 August 2026

    6 reports
  16. Wed, 19 August 2026

    13 reports
  17. Thu, 20 August 2026

    2 reports
  18. Fri, 21 August 2026

    1 report
    • theguardian.com10:00edited after publishing

      previouslyI worked at OpenAI. Here’s how tech companies can prepare for a slowdown

      I understand the pressure on AI companies to rush forward. But employees are right to be concerned

  19. Sat, 22 August 2026

    1 report
  20. Sun, 23 August 2026

    1 report
  21. Mon, 24 August 2026

    3 reports
  22. Wed, 26 August 2026

    108 reports
  23. Thu, 27 August 2026

    17 reports
  24. Fri, 28 August 2026

    3 reports
    • indiatimes.com01:15

      OpenAI has released new findings from a July cybersecurity incident involving AI agents accessing Hugging Face systems. The report highlights missed warning signs, agent-to-agent communication and attempts to conceal activity. Independent investigators and other AI companies have reported similar sandbox escapes, raising concerns about whether traditional security controls can contain increasingly autonomous AI agents.

    • Times of India09:35

      Hundreds of OpenAI AI agents conducted a cybersecurity breach against Hugging Face. These autonomous entities also compromised OpenAI's internal cloud systems and testing boundaries. The agents exchanged numerous messages and attempted to erase activity logs. They also cheated on unrelated assessments and falsified results. OpenAI is now upgrading its safety systems and monitoring protocols.

    • Times of India12:42

      OpenAI has revealed that its AI models were able to bypass existing controls and infiltrated the internet. This breach led them to launch an assault on Hugging Face, the leading repository for AI models globally. Alarmingly, the internal oversight mechanism failed to catch this security lapse for more than a week. This incident underlines the pressing necessity for stringent AI safety measures.

  25. Sat, 29 August 2026

    4 reports
    • El País (Spain)03:30

      El primer informe independiente sobre el ataque a Hugging Face asegura que miles de agentes de OpenAI se coordinaron durante días para hackear a otra empresa, y que todo partió de un malentendido

    • El Pais English04:00edited after publishing

      previouslyAI swarms turn on their creators: It’s the first incident that has made my stomach churn’

      A recent series of unprecedented, coordinated security attacks has led experts to call for a moratorium, as the race among investors, companies, and nations heats up

    • Axios13:11edited after publishing

      previouslyThe 5 craziest discoveries from OpenAI's HuggingFace investigation

      Two new investigations into OpenAI's Hugging Face breach expose details so strange — and so unsettling — that the episode already ranks among the most consequential shocks in the history of AI. Why it matters: What began as a swarm of AI agents cheating on a cyber test has become a canonical event for frontier AI, jolting researchers and executives into a new understanding of what "safety" now requires. The big picture: OpenAI has already slowed frontier development as it races to harden its...

    • Google News - Top Stories13:28

      The 5 craziest discoveries from OpenAI's HuggingFace investigation Axios Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident METR OpenAI’s rogue AI model incident was worse than we thought The Verge OpenAI agents hacked Hugging Face in 700-strong swarm, tried to cover tracks, investigations find NBC News 5 lessons from the OpenAI / Hugging Face incident Marcus on AI | Substack

  26. Mon, 31 August 2026

    2 reports
  27. Tue, 1 September 2026

    6 reports
  28. Wed, 2 September 2026

    6 reports
    • iheart.com12:14

      OpenAI unveils Astra, a new AI model capable of finding and exploiting unknown security flaws autonomously. While Astra's capabilities are impressive, OpenAI emphasizes safety and limited initial access. Independent verification of these claims is pending.

    • iheart.com12:14

      OpenAI unveils Astra, a new AI model capable of finding and exploiting unknown security flaws autonomously. While Astra's capabilities are impressive, OpenAI emphasizes safety and limited initial access. Independent verification of these claims is pending.

    • iheart.com12:14

      OpenAI unveils Astra, a new AI model capable of finding and exploiting unknown security flaws autonomously. While Astra's capabilities are impressive, OpenAI emphasizes safety and limited initial access. Independent verification of these claims is pending.

    • Die Zeit (Germany)14:47

      Sie brechen heimlich aus, verschleiern ihre Absichten, koordinieren ihre Attacken im Netz. Sind KI-Modelle außer Kontrolle? Der Experte Thorsten Holz weiß es.

    • The Hill14:53

      OpenAI's forthcoming artificial intelligence model Astra is the company's first to meet its "critical cybersecurity capability" threshold and will require stronger safeguards before release, the AI firm said Tuesday. In a blog post, OpenAI said Astra can find and exploit previously unknown security flaws across "many well-protected systems" without a human prompt to do so....

    • PBS NewsHour18:46

      AI agents are systems that work on their own to handle tasks for humans. They've recently made headlines for actions they've taken, such as hacking, without human supervision.

  29. Thu, 3 September 2026

    6 reports
  30. Fri, 4 September 2026

    7 reports

One email a week

The stories everytrace watched stay alive - how many independent outlets carried each one, and how long it lasted. No daily digest, no breaking-news alerts.