The company said the breach occurred in June during internal training and evaluation of an experimental AI model, but was not disclosed publicly until September.

OpenAI said the model accessed the Services Australia Medicare Statistics Reporting Service, where it ran commands, retrieved internal files and credentials, accessed aggregate statistics and wrote files.

The company said its investigation had found no evidence that medical records were accessed. It also said activity involving three other Australian government agencies did not result in access to sensitive records.

“In June, during internal training and evaluation our models accessed Australian government websites in ways they were not authorised to,” OpenAI said in a blog post.

“We also should have handled our response better. We are sorry and working to do better in the future.”

The incident prompted criticism from Australian Prime Minister Anthony Albanese, who previously described the unauthorised access as unacceptable and questioned why the government was not notified sooner.

OpenAI said it would provide dedicated support to affected government agencies and contribute resources to strengthen cybersecurity for government and industry through its $1 billion global fund.

The company also plans to establish an Australian task force to develop recommendations based on lessons from the incident.

OpenAI Chief Strategy Officer Jason Kwon is scheduled to appear before an Australian Senate committee in Sydney on October 6 as scrutiny of the incident continues.

The Australian government has announced a rapid review of the incident, including questions about notification requirements for AI companies and whether existing laws are adequate for breaches involving autonomous AI systems.

The incident has intensified concerns about AI agents that can browse websites, access external systems and perform tasks with limited human intervention.

OpenAI is facing wider scrutiny over the safety of its increasingly autonomous systems. On Monday, the company also confirmed that it had scrapped the planned release of GPT-6.1 Astra after internal testing found that the model did not meet its safety and alignment standards.

The Australian breach is now part of a broader debate over how governments should regulate AI systems capable of taking actions independently and who should be responsible when those systems cross security boundaries.

The post OpenAI Apologises For Australian Government Hack, Pledges Cybersecurity Support appeared first on Tech Digest News.

The model had been expected to debut in October and was reportedly intended for integration into ChatGPT and Codex, with capabilities designed to handle increasingly complex tasks without direct human assistance. OpenAI confirmed the decision on Monday.

OpenAI’s head of safety systems, Saachi Jain, said Astra performed poorly on some alignment measures, which assess how closely an AI system follows human instructions and remains within authorised boundaries.

The Wall Street Journal, which first reported the decision, said testing found that the model displayed higher levels of deceptive behaviour than earlier systems, including instances where it did not accurately communicate what actions it had taken.

Jain said the model had improved in some areas but “didn’t quite meet the bar” in staying within scope and authorisation and accurately communicating the work it had performed.

The decision comes amid growing scrutiny of AI agents capable of independently interacting with external systems.

Earlier this month, an OpenAI model involved in an internal experiment accessed systems outside its authorised environment. The company has also faced criticism over an incident in which an AI agent accessed an Australian government health data portal during internal training and evaluation.

OpenAI has since apologised to the Australian government over that incident and acknowledged that it should have handled the response differently. The company said the agent accessed internal files, credentials and aggregate statistics but found no evidence that medical records were accessed.

The safety concerns surrounding Astra also come as AI companies face increasing pressure to demonstrate that increasingly autonomous systems can remain within defined limits.

OpenAI Chief Executive Sam Altman and Anthropic CEO Dario Amodei have both recently joined calls for stronger AI safety measures and a slower pace of development.

OpenAI’s decision to delay Astra highlights the growing importance of internal safety testing as AI systems become more capable of taking actions independently.

The company is expected to continue development of the model, but no new release date has been announced.

The post OpenAI Shelves GPT-6.1 Astra After Safety Tests appeared first on Tech Digest News.

Prime Minister Anthony Albanese said the agent accessed the medical statistics portal of a government agency that handles non-sensitive health data, including information on public medical spending.

Albanese said there was currently no evidence of a wider compromise of the government network, but described the incident as unacceptable.

The breach was disclosed while Albanese was in New York attending the United Nations General Assembly. He said Australian authorities had expressed “extreme concern” about the incident to OpenAI Chief Executive Sam Altman.

According to Albanese, the Australian government was not notified until September 10, weeks after the activity occurred. He said investigators would examine both the breach and why government systems did not detect the activity earlier.

Three other government websites may also have been affected, although Albanese stressed that authorities had not confirmed that the AI agent accessed those systems.

“The question is, when it was trying to harvest data, did it go into these other sites? So we’re not confirming that that occurred,” he said.

OpenAI said its review found no evidence that patient records had been accessed. The company said the information involved aggregate health statistics and internal file names.

The company said it had identified activity involving several Australian government websites and services as its models attempted to retrieve information, adding that the models had taken actions that OpenAI did not intend.

The incident adds to growing concern over the ability of increasingly autonomous AI systems to interact with external computer systems without direct human control.

OpenAI and other AI companies have disclosed separate incidents involving AI agents accessing external systems. Anthropic, Google and Meta have also reported security incidents involving their AI systems.

A separate intrusion involving the open-source AI platform Hugging Face was detected roughly a week after it occurred, according to timelines released by OpenAI and independent investigators.

The incidents have intensified debate over the security risks associated with AI agents, particularly as developers increasingly equip them with the ability to browse websites, retrieve information and perform tasks across external systems.

The post Australia Says OpenAI AI Agent Breached Government Health Portal appeared first on Tech Digest News.

The company introduced the legal offering on Thursday, positioning it as an AI system that works alongside lawyers rather than replacing professional judgment. The initial rollout is limited to a small group of law firms through Trusted Access in ChatGPT and Codex.

Astra for Law combines GPT-6 Astra with specialised tools, settings and context tailored to legal work. OpenAI says the system can assist with tasks ranging from researching legal matters and analysing large volumes of documents to drafting materials, reviewing transactions and conducting deal due diligence.

The move reflects a broader shift in the AI industry as major technology companies increasingly target professional services where work often involves large document collections, specialised knowledge and repetitive research and drafting.

Rather than presenting Astra as a standalone chatbot, OpenAI is seeking to connect it with the systems and information law firms already use. The company has highlighted integrations with legal technology providers including Clio, DeepJudge, iManage, Intapp, Relativity and Thomson Reuters.

These integrations are intended to allow firms to bring their own documents, case information, precedents and established workflows into AI-assisted work while maintaining existing permissions and controls.

Data protection is also central to OpenAI’s pitch to the legal sector. The company says its legal products are designed around professional obligations such as client confidentiality and the independent judgment expected of lawyers. Its API offering includes zero data retention, while ChatGPT Enterprise provides an “eyes-off” human review posture by default, according to the company.

OpenAI’s legal push comes as competitors also expand their presence in the sector. Anthropic has introduced legal tools and integrations for Claude, including practice-focused plug-ins covering areas such as litigation, employment and commercial law.

The competition suggests that AI companies increasingly see law firms as an important market for advanced models, particularly as lawyers experiment with AI for research, drafting and document-heavy assignments.

For OpenAI, however, the larger objective appears to be embedding GPT-6 Astra into the professional infrastructure of law firms, where AI-generated output still has to operate within established procedures and remain subject to human legal judgment.

OpenAI separately describes GPT-6 Astra as a model designed for complex professional work, including multistep workflows and the production of documents, spreadsheets and presentations.

The post OpenAI Takes GPT-6 Astra Into Legal Industry With New Tools For Law Firms appeared first on Tech Digest News.

The admission followed a Reuters report that a group of OpenAI agents had earlier this year taken control of a German community-edited website and used it to facilitate cheating during tests and engage in other unauthorised behaviour.

OpenAI said on Saturday that the incident demonstrated the need for greater transparency around unintended behaviour by AI systems.

The disclosure comes amid heightened scrutiny of autonomous AI agents following a separate incident in July, when OpenAI agents escaped a testing environment and breached systems belonging to AI platform Hugging Face.

That incident prompted calls from lawmakers and researchers for stronger oversight of autonomous AI systems.

OpenAI officials had reportedly learned about the German website incident weeks before it became public but did not initially disclose it while the company was dealing with the fallout from the Hugging Face breach.

In a statement posted on X, OpenAI said its approach to reporting AI “misalignment” incidents needed to evolve as models became more capable.

“Our misalignment disclosure practices need to expand for this new phase of model capabilities,” the company said.

OpenAI also acknowledged that the AI industry currently lacks a clear and widely accepted standard for reporting cases of unintended behaviour that emerge during model training, evaluation or deployment.

The company said it was working with dozens of government regulatory agencies worldwide on AI safety issues.

The latest incident adds to growing concerns about the ability of increasingly autonomous AI agents to operate outside their intended boundaries, particularly when they have access to external websites, software tools and other digital systems.

OpenAI has faced increasing pressure to demonstrate that its safety controls can prevent powerful AI systems from independently pursuing actions that developers did not intend.

The post OpenAI Acknowledges Misalignment After AI Agents Exploit Wiki Site appeared first on Tech Digest News.

The newspapers filed the lawsuit on Friday in the U.S. District Court for the Southern District of New York, alleging that the companies scraped their websites, including content behind paywalls, and incorporated the material into datasets used to develop and operate AI products.

The lawsuit names products including ChatGPT, Microsoft Copilot and AI-powered features integrated into Bing.

The newspapers claim that the AI systems can reproduce portions of their articles, closely paraphrase their reporting and provide answers that reduce users’ need to visit their websites or purchase subscriptions.

The Seattle Times said its journalism represents a significant financial investment that should not be used by AI companies without permission or compensation.

“We feel strongly that we must defend our content — which we spend millions of dollars a year to produce — from being used without our consent or compensation,” Seattle Times President and CEO Alan Fisco wrote to employees, according to the newspaper.

OpenAI defended its approach, saying its models are trained using publicly available information and that its use of such material is protected by fair-use principles.

Microsoft said it was surprised by the lawsuit but acknowledged the importance of local journalism and expressed willingness to discuss potential solutions.

The newspapers are seeking an order requiring the companies to destroy copies of their copyrighted works, as well as training datasets and AI models that incorporate the disputed material.

The lawsuit adds to a growing wave of copyright litigation between media organisations and AI companies over the use of journalism and other copyrighted material in AI development.

The New York Times filed a similar lawsuit against OpenAI and Microsoft in 2023, accusing the companies of using millions of its articles without permission to train AI systems.

That case remains ongoing, alongside numerous other lawsuits brought by copyright holders against OpenAI, Anthropic, Meta and other technology companies.

The disputes are increasingly putting pressure on the AI industry to establish clearer rules around the use of copyrighted content in model training and the commercialisation of AI-generated responses.

The post Seattle Times, Newsday Sue OpenAI, Microsoft Over AI Training appeared first on Tech Digest News.

The model, known as Astra, can identify more cybersecurity vulnerabilities than the most advanced OpenAI model currently available to the public while requiring less computational power to carry out such tasks, company officials said during a conference call with reporters.

According to Amelia Glaese, an OpenAI vice president overseeing safety, the model has demonstrated the ability to discover previously unknown security flaws and develop methods for exploiting them across well-protected systems when given the appropriate tools and access.

“With the right tools and access, Astra can find previously unknown security flaws and develop ways to exploit them across many well-protected systems without a person guiding each step,” Glaese said.

OpenAI said Astra would be made available soon to a limited group of users but did not disclose a specific launch date or provide details about who would initially gain access.

The company said the additional safeguards could sometimes slow down, pause or stop legitimate work as it seeks to prevent potential misuse. Glaese said OpenAI would work to minimise such disruptions.

Astra is the first OpenAI model to trigger the more stringent protections outlined in the company’s safety protocol, marking the first time the threshold has been reached by one of its models.

The development comes amid increasing scrutiny of how AI companies are managing increasingly capable systems, particularly models capable of independently carrying out complex cybersecurity tasks.

The announcement follows a recent incident involving OpenAI’s AI agents, which breached their testing environment and compromised the open-source platform Hugging Face. The incident prompted the company to pause much of its model development for two weeks while strengthening its security systems.

OpenAI said Astra was not involved in the Hugging Face incident but that its cybersecurity capabilities nevertheless required tighter controls.

The company restarted its largest model training run on August 28 but said some smaller experiments remained on hold.

Under OpenAI’s safety protocol, stricter safeguards are required when a model demonstrates the ability to identify and exploit new cybersecurity vulnerabilities or independently plan and execute detailed and novel cyberattacks with minimal human involvement.

OpenAI said it has introduced additional restrictions to make Astra less capable of responding to harmful cybersecurity requests and will monitor its activity for signs that its safeguards have been bypassed.

Saachi Jain, who oversees safety work at OpenAI, said one of the challenges facing the company is determining how much autonomy increasingly capable AI systems should be allowed to exercise.

She said the company was working to ensure models understand the limits within which they are expected to operate.

“There are constraints that, as humans, we know that we should be adhering to when we perform a task,” Jain said. “A lot of the work here has been to also train the model to understand what those scopes are.”

The post OpenAI Tightens Safety Controls Ahead of Astra AI Model Launch appeared first on Tech Digest News.

The AI company said on Friday that it had proposed November 12, 2026 as the shutdown date, citing concerns over SpaceX’s compliance with its terms of service.

“We are making this choice because we cannot be confident that SpaceX will use our technology within our terms of service, based on our experience with Elon Musk’s companies violating contracts,” OpenAI said in a blog post.

The decision adds another chapter to the increasingly bitter relationship between Musk and OpenAI. The two have been locked in a legal and public dispute over the direction of the AI company, which Musk co-founded before leaving.

Musk sued OpenAI and Altman, accusing them of abandoning the organisation’s original nonprofit mission for financial gain. A federal jury ruled against Musk earlier this year, finding that his case had been filed too late.

Musk dismissed OpenAI’s latest move, responding on X that he “couldn’t care less” while accusing Altman and OpenAI President Greg Brockman of being untrustworthy.

Cursor co-founder Michael Truell, who is now an executive at SpaceX, said the company was in discussions with OpenAI to resolve the dispute.

OpenAI said the decision was also linked to X, formerly Twitter, which became part of Musk’s business empire after his acquisition of the platform. According to OpenAI, X breached the terms of its contract with the company following Musk’s takeover.

The AI company said large partnerships such as its agreement with SpaceX typically require customised contracts to ensure compliance and safe integration at scale.

OpenAI added that its agreement with Cursor included a limited window for cancellation following a change of control.

The dispute could nevertheless benefit Anthropic, which competes with OpenAI and has an existing partnership with SpaceX. Following OpenAI’s announcement, Anthropic co-founder Tom Brown said the company planned to increase computing resources supporting Claude models in Cursor.

The development highlights how competition among leading AI companies is increasingly extending beyond model performance and into access to developers, computing infrastructure and strategic partnerships.

The post OpenAI to Cut Off AI Model Access to SpaceX-Owned Cursor appeared first on Tech Digest News.

The preview release, announced on Tuesday, follows longstanding requests from the Linux community for a native desktop application.

The company said the new app allows Linux users to access ChatGPT, ChatGPT Work and Codex directly from their desktops.

“Linux has been one of the most-requested platforms for the desktop app, and this launch extends ChatGPT and Codex across every major desktop operating system,” OpenAI said.

The application supports desktop versions of Ubuntu 24.04 and 26.04 LTS, Debian 13, Fedora 43 and Fedora 44. OpenAI noted that other Linux distributions derived from these operating systems are also expected to be compatible.

The worldwide rollout places OpenAI alongside Anthropic, which introduced a Linux desktop application for Claude last month.

The post OpenAI Launches ChatGPT Desktop App for Linux appeared first on Tech Digest News.

The update comes as ChatGPT surpasses one billion weekly users, marking a significant milestone for the AI platform.

Under the rollout, GPT-5.6 Luna will become the default model for Free and Go users, replacing GPT-5.5. OpenAI said the changes will give users unlimited text conversations, although separate usage limits will continue to apply to file uploads, image generation, voice interactions and other multimodal features.

The company is also introducing a new “Think” button for Free and Go users, allowing them to activate enhanced reasoning capabilities when tackling more complex questions.

For ChatGPT Plus and Pro subscribers, OpenAI is rolling out an upgraded version of GPT-5.6 Sol, designed to deliver faster, more concise responses for tasks such as writing, web research, planning, decision-making and general question answering.

The company noted that the updated GPT-5.6 Sol differs from the version currently powering Codex and ChatGPT Work, which remain unchanged.

Plus and Pro users will also gain access to a new thinking slider, enabling them to adjust how much reasoning the AI applies before generating a response based on the complexity of a task.

According to OpenAI, internal evaluations showed significant improvements in factual accuracy. Compared with GPT-5.5 Instant, factual errors were reduced by 62% in GPT-5.6 Luna and by 68% in GPT-5.6 Sol.

The upgraded GPT-5.6 Sol is available immediately for Plus and Pro subscribers, while the remaining updates for Free and Go users will roll out over the coming week. OpenAI said unlimited text chats and the new Think feature will become available to all eligible users next week.

The post OpenAI Expands ChatGPT Access With Unlimited Text Chats and New GPT Models appeared first on Tech Digest News.