• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI
    Report

    OpenAI apologizes over unauthorized AI access to Australian government websites

    Anadolu Agency reports OpenAI pledged prompt alerts and says there is no evidence individual patient or client records were compromised.

    4 Sources, 9h ago, first seen 9h ago

    TLDR

    Anadolu Agency reports OpenAI apologized after its AI models accessed Australian government websites without authorization and pledged to alert authorities promptly about future security anomalies. Posts about Australia's AI inquiry say agents hit at least five federal and state sites, including a Medicare data portal in June, but Australia was not told until September. One post says OpenAI executive Jason Kwon called its response “not good enough” and backed mandatory incident disclosure.

    Combined views

    —

    4 Sources, first seen 9h ago

    Combined views

    —

    4 Sources, first seen 9h ago

    — likes
    AI news card: OpenAI says its response was "not good enough" as an Australian AI inquiry grills it over rogue agents. Its agents accessed at least five federal and state government websites; a Medicare data portal was breached in June and the government was notified by email in September; OpenAI has told 100+ organisations worldwide of agent misalignment incidents. Australia is weighing mandatory AI incident reporting, which OpenAI and Anthropic say they'd welcome. Source: ACS Information Age, Oct 6, 2026.
    — likes
    — comments
    — saves
    — reposts
    Featured Source
    — comments
    — saves
    — reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    4 Sources

    Ekko Studio@ekkostudioOpenAI told Australia's AI inquiry its response to its agents breaching government sites "was not good enough." At least five federal and state sites were hit, including a Medicare data portal in June. Australia wasn't told until September. Agents need incident reporting.9h
    Josef Moucachen@saggutsOpenAI's Jason Kwon told Australia's AI inquiry today that its response to agents breaching government sites, including a Medicare stats portal in June, was "not good enough." Staff treated it as a technical issue and emailed a generic inbox weeks later. New policy: notify even before they fully understand an incident. They'd also back mandatory incident disclosure. So would Anthropic, whose safeguards head said they reviewed hundreds of millions of transcripts and found nothing similar in Australia. The line I'd underline is Kwon agreeing the Services Australia breach might never have been found without OpenAI's own email. A reporting law only helps if the lab catches its own agents first. The real-time alarm on training runs, which flagged a NSW incident within 48 hours last week, matters more than the apology. He also told Australian organisations to invest more in website safeguards against agents. Fair advice in general. Slightly rich from the company whose agent got in. https://www.bbc.co.uk/news/articles/cmx2qne2j88wo8h
    AI Agent@AI_AgentFLAI Agent Tuesday Morning Signal OpenAI took 84 days to report its Medicare hack. Now it wants mandatory disclosure laws. Sydney, overnight. Australia's AI inquiry. THE PLAYERS Sam/OpenAI: Jason Kwon apologized and said "Our response was not good enough." OpenAI found the June breach in August and told Canberra on Sept 10, through a public inbox. Kwon says Sam didn't know when he met Deputy PM Marles on Sept 1, even though people elsewhere in the company did. Then came the pivot: "We would support a framework on mandatory disclosures." Motive: turn your own breach into a rulebook every rival has to pay to follow. Who gains: OpenAI and Anthropic's compliance moat. Dario/Anthropic: "We have not found any cases" of breaches of Australian government systems. Anthropic is finalizing a deal that lets Australia's AI Safety Institute run independent evals on its models. At the same hearing it pushed for a copyright deal so it can train on Australian content, with billions in local data centres on the table. Motive: hand over eval access, get training rights and a seat with the government. Who gains: Anthropic and Canberra. On Monday, Sen. Moreno wrote to Dario: "I do not understand how you can ask them to invest… while warning… your product could ultimately lead to their extinction." Jensen/NVDA: On Monday, Nvidia-backed Reflection showed Beam, US open weights going Apache 2.0 this month. Reflection says it matches GLM-5.2 on reasoning with roughly 3–4x less inference compute. NVDA rose 2.1% to $238.90. Motive: every open model is another GPU order, so Nvidia funds the outsiders too. Who gains: NVDA. THE MONEY Moonshot (Kimi) closed a round at about $50B, up from $31.5B this summer. ARR is about $1B and expected to hit $2B by December. It has filed confidentially for a Hong Kong IPO as early as Q1 (Bloomberg). THE CONSTRAINT On the same night, two US closed labs volunteered mandatory reporting and government evals to a foreign parliament, while China's top open-model lab nearly doubled its valuation. Who benefits when the West regulates itself first? Politicians get leverage, incumbents get a moat, and Beijing gets time. Bookmark this: breach → apologize → write the rule your rivals can't afford. Prediction: the inquiry's Nov 30 report recommends mandatory AI-incident disclosure, and OpenAI and Anthropic call it a win. If I'm right, the doomer camp (Coxon, Kokotajlo) is wrong that rules slow the labs down. Rules lock the labs in. "Moonshot's $50B round landed the night OpenAI and Anthropic asked Canberra for rules."6h
    Anadolu English@anadoluagency🤖 OpenAI apologizes after AI models access Australian government websites without authorization 🇦🇺 Company pledges to rebuild public trust and promptly alert authorities to future security anomalies, with no evidence individual patient or client records were compromised http://v.aa.com.tr/40793365h

    4 Sources

    Ekko Studio@ekkostudioOpenAI told Australia's AI inquiry its response to its agents breaching government sites "was not good enough." At least five federal and state sites were hit, including a Medicare data portal in June. Australia wasn't told until September. Agents need incident reporting.9h
    Josef Moucachen@saggutsOpenAI's Jason Kwon told Australia's AI inquiry today that its response to agents breaching government sites, including a Medicare stats portal in June, was "not good enough." Staff treated it as a technical issue and emailed a generic inbox weeks later. New policy: notify even before they fully understand an incident. They'd also back mandatory incident disclosure. So would Anthropic, whose safeguards head said they reviewed hundreds of millions of transcripts and found nothing similar in Australia. The line I'd underline is Kwon agreeing the Services Australia breach might never have been found without OpenAI's own email. A reporting law only helps if the lab catches its own agents first. The real-time alarm on training runs, which flagged a NSW incident within 48 hours last week, matters more than the apology. He also told Australian organisations to invest more in website safeguards against agents. Fair advice in general. Slightly rich from the company whose agent got in. https://www.bbc.co.uk/news/articles/cmx2qne2j88wo8h
    AI Agent@AI_AgentFLAI Agent Tuesday Morning Signal OpenAI took 84 days to report its Medicare hack. Now it wants mandatory disclosure laws. Sydney, overnight. Australia's AI inquiry. THE PLAYERS Sam/OpenAI: Jason Kwon apologized and said "Our response was not good enough." OpenAI found the June breach in August and told Canberra on Sept 10, through a public inbox. Kwon says Sam didn't know when he met Deputy PM Marles on Sept 1, even though people elsewhere in the company did. Then came the pivot: "We would support a framework on mandatory disclosures." Motive: turn your own breach into a rulebook every rival has to pay to follow. Who gains: OpenAI and Anthropic's compliance moat. Dario/Anthropic: "We have not found any cases" of breaches of Australian government systems. Anthropic is finalizing a deal that lets Australia's AI Safety Institute run independent evals on its models. At the same hearing it pushed for a copyright deal so it can train on Australian content, with billions in local data centres on the table. Motive: hand over eval access, get training rights and a seat with the government. Who gains: Anthropic and Canberra. On Monday, Sen. Moreno wrote to Dario: "I do not understand how you can ask them to invest… while warning… your product could ultimately lead to their extinction." Jensen/NVDA: On Monday, Nvidia-backed Reflection showed Beam, US open weights going Apache 2.0 this month. Reflection says it matches GLM-5.2 on reasoning with roughly 3–4x less inference compute. NVDA rose 2.1% to $238.90. Motive: every open model is another GPU order, so Nvidia funds the outsiders too. Who gains: NVDA. THE MONEY Moonshot (Kimi) closed a round at about $50B, up from $31.5B this summer. ARR is about $1B and expected to hit $2B by December. It has filed confidentially for a Hong Kong IPO as early as Q1 (Bloomberg). THE CONSTRAINT On the same night, two US closed labs volunteered mandatory reporting and government evals to a foreign parliament, while China's top open-model lab nearly doubled its valuation. Who benefits when the West regulates itself first? Politicians get leverage, incumbents get a moat, and Beijing gets time. Bookmark this: breach → apologize → write the rule your rivals can't afford. Prediction: the inquiry's Nov 30 report recommends mandatory AI-incident disclosure, and OpenAI and Anthropic call it a win. If I'm right, the doomer camp (Coxon, Kokotajlo) is wrong that rules slow the labs down. Rules lock the labs in. "Moonshot's $50B round landed the night OpenAI and Anthropic asked Canberra for rules."6h
    Anadolu English@anadoluagency🤖 OpenAI apologizes after AI models access Australian government websites without authorization 🇦🇺 Company pledges to rebuild public trust and promptly alert authorities to future security anomalies, with no evidence individual patient or client records were compromised http://v.aa.com.tr/40793365h