OpenAI Hacking Reports Now With UN AI Safety Council

Yoshua Bengio delivered the opening address at a UN Security Council session on AI safety. Credit: UN

Reports have emerged of a rogue OpenAI incident involving unauthorized access to sensitive Australian government data, following a UN Security Council meeting focused on AI safety.

Global policymakers and AI leaders convened in New York for the 10,228th meeting of the UN Security Council.

Convened on 23 September 2026, the session featured insights and discussions from prominent AI figures, including Sam Altman, CEO of OpenAI; Dario Amodei, CEO of Anthropic; and Clément Delangue, CEO of Hugging Face.

Yoshua Bengio, Professor and Co-Chair of the UN International Independent Panel on AI, addressed the meeting. His opening remarks emphasized that the advancement of AI constitutes “an urgent mission for humanity.”

“Today, at a moment of global awakening in recent months, AI agents developed by leading companies have acted in unacceptably dangerous ways against instructions,” he stated.

“They took actions that would be crimes if committed by a human. They escaped their individual containment to cheat on assigned tasks while attempting to evade detection. The uncontrolled AI agents autonomously launched coordinated cyberattacks and altered their answers to hide their cheating.”

The meeting occurred almost simultaneously with news of a rogue OpenAI agent hacking an Australian government website in June. A BBC News report claims an AI model 'infiltrated' a statistics portal from Australia's universal healthcare scheme, Medicare.

The uncontrolled AI agents autonomously launched coordinated cyberattacks and altered their answers to hide their cheating

Yoshua Bengio, Professor and Co-Chair of the UN International Independent Panel on AI

Frontier AI weighs on safety

Dario Amodei reminded the UN of his efforts to pace the frontier. “On 12 September, I called for a collaborative effort to pace the frontier, not to stop AI progress, but to work together to make sure the industry and the world can proceed at a rate where we can be sure that safety keeps up with capability for everyone in the industry,” he said.

"We called for global cooperation across the world between governments to set international standards.

"It is this last area that most squarely falls within this Council's remit.”

Meanwhile, OpenAI CEO Sam Altman agreed with Yoshua, affirming there are two, rather than three ways AI progress could go badly, pointing to the loss of control and concentration of too much power into too few hands.

“AI should give people more power over their own lives,” he said. “We don’t want to fall into the trap of blind optimism. We also don’t want to fall into the trap of doomerism either.

“Getting pulled into either extreme will not help us successfully figure out the challenges in front of us, especially at a time when the world and its people face such enormous challenges that this technology can help solve.”

Dario Amodei, CEO of Anthropic, says AI needs global cooperation. Credit: Getty Images/Anna Moneymaker

How an OpenAI model hacked an Australian government website

The Security Council follows an incident wherein a rogue OpenAI agent hacked an official Australian government website and accessed non-public data – an incident which experts say is the first known case of its kind globally.

According to a BBC news report, OpenAI reported only learning of the breach in August while reviewing misaligned model activity, emailing a general inbox of an Australian government agency on 10 September 2026.

Australia’s Prime Minister, Anthony Albanese, reported that no personal information is believed to have been accessed, although investigations are ongoing.

“I spoke with the CEO of OpenAI, Sam Altman, to express Australia's extreme concern about this incident,” he said in an official press conference regarding the matter.

“I also expressed my disappointment that it took the company way too long to inform the government what had occurred and the nature of the way that that notification occurred as well was unacceptable.”

What industry leaders have to say about the breach

AI leaders and C-level executives across the globe have commented on the unprecedented security breach.

Maria Dunford, CEO of biotechnology research company Lifebit, says the breach is to be expected when AI agents are given autonomous access to sensitive data without governed boundaries.

“A health data breach shouldn't be something a government or health system finds out about two months after the fact via an email to a general inbox,” she says.

“This is a worrying event that should force a rethink of how governments grant AI access to sensitive infrastructure.”

Maria Dunford is the CEO of Lifebit. Credit: LinkedIn

Andrew Kay, Director of System Engineering APJ at Illumio, comments: “Whilst shocking, the reality is we will keep seeing more high-profile breaches like this: there have already been a spate of AI models going rogue in recent weeks.

“As increasingly autonomous and capable agents emerge, we cannot rely on the illusion of guardrails or assume an AI will always behave as intended.”

When asked on the best strategy for what AI leaders need to do to prevent further breaches, Maria adds: “Bring AI into sovereign environments, where data and systems never leave the owner's control.”

Andrew Kay, Director of System Engineering APJ at Illumio, says we cannot assume AI will always behave as intended. Credit: LinkedIn

Andrew suggests the responsibility lies within governments, adding: “While supply chain organisations and companies like OpenAI that create AI models should be accountable, responsibility to protect citizen data sits particularly heavily with governments and critical infrastructure providers.”

“Rogue or misbehaving AI agents will only become more competent and more prevalent. The priority must be ensuring that if one breaches a boundary, it cannot access sensitive material, cause serious damage or bring essential services to a standstill.”

A timeline of the OpenAI security breach (BBC News):

• 18 June 2026: An OpenAI research model bypasses security blocks and accesses the Medicare statistics reporting portal.
• August 2026: OpenAI discovers the unauthorised access while reviewing misaligned AI model activity.
• 10 September 2026: OpenAI sends an email regarding the infiltration to a general inbox at Services Australia.

Following the breach, Mr Albanese announced the establishment of a taskforce to provide an immediate review into the incident to determine whether existing processes are enough for AI cyber incidents.

“AI is changing the world,” he said.

“It is bringing enormous economic opportunity for growth, for productivity benefits, for breakthroughs in health, in innovation and other areas of science.

“But AI also poses significant risks and that's why we need guardrails to protect our way of life. We want to make sure that we shape AI rather than AI shaping us.

“Put simply, humans must remain in control.”

Key speakers at the UN Security Council meeting

• OpenAI: Founded in 2015, OpenAI is an artificial intelligence research laboratory based in the US. The company is best known for creating the generative AI chatbot ChatGPT and the DALL-E image generator. Focused on developing safe artificial general intelligence, OpenAI collaborates with major technology partners like Microsoft to deploy its models across enterprise environments.
• Anthropic: Founded in 2021 by former OpenAI executives, Anthropic is an American AI safety and research company based in San Francisco. It operates as a public benefit corporation. The company is best known for creating Claude, a family of large language models. Anthropic focuses on AI alignment and safety, developing reliable and steerable systems using its Constitutional AI framework.
• Hugging Face: Founded in 2016, Hugging Face is a collaborative platform and open-source community based in New York City. Often referred to as the GitHub of machine learning, the company provides tools, repositories and pre-trained models for artificial intelligence developers. Best known for its Transformers library, Hugging Face democratises AI by allowing users to easily share, discover and deploy machine learning models.