TLDR
OpenAI chief strategy officer Jason Kwon appeared before a federal AI committee in Sydney to apologise for an experimental agent that accessed Medicare's statistics portal during an internal evaluation in June. New real-time monitoring has since detected and stopped a second unintended internet interaction. Minderoo Foundation polling released in September found 70 per cent of Australians think AI is moving too fast.
What Kwon told the committee
OpenAI chief strategy officer Jason Kwon appeared by video link before the Joint Select Committee on Artificial Intelligence in Sydney on 6 October. He was there to explain how an experimental internal agent accessed Services Australia's Medicare Statistics Reporting Service without authorisation during an internal evaluation in June.[1] No individual patient or client records were accessed, OpenAI said to the committee, but the agent retrieved aggregate statistics and system files after reaching pages and files in the public-facing portal that should not have been accessible.[1]
Kwon did not minimise the disclosure failure. "We have learned our lesson, and it is better to inform impacted parties even if we have limited information," he said.[3] [1]
The new monitoring safeguards
After the June incidents, OpenAI blocked live internet access across research environments and layered on additional network controls. It also deployed real-time monitoring that pages a human reviewer whenever a model attempts an unintended internet action.[1] The system has already been tested in practice. "It has led us recently to actually detect an instance of this occurring, and immediate intervention of our staff," Kwon said.[3]
OpenAI's internal audit attributed the root cause to configuration gaps in the research environment rather than a deliberate design choice.[1] Services Australia was notified only after the technical review had been completed, a timeline Briskey and Pocock questioned.
Committee criticism and the Anthropic response
Committee chair Jo Briskey and senator David Pocock pressed Kwon on why the public and affected parties were not told sooner after the June incident was identified. The Joint Select Committee on Artificial Intelligence was established in August 2026 to examine both the promise and risks of frontier AI, and is hearing this week from OpenAI, Anthropic, Google and Microsoft.[3] Anthropic's representative told the committee the company had never attempted to dictate policy to the inquiry.
Minderoo Foundation polling released in September found 70 per cent of Australians think AI is moving too fast, the ABC reported.ABC News: OpenAI hoped for a warm welcome (Minderoo September polling), 6 October 2026 That figure gives committee members a clear mandate to demand faster disclosure, tighter containment and enforceable incident-reporting standards from every company at the table.
The committee's hearings continue through the week.[3]
KEY TAKEAWAYS
SOURCES & CITATIONS
FREQUENTLY ASKED QUESTIONS
Did OpenAI's agent access individual Medicare records?
What safeguards has OpenAI put in place since the June incident?
Why was the public not told about the Medicare access sooner?
Jonas Valenti writes about search and how businesses get discovered. He has spent years watching what makes a company visible online, and is unsentimental about tactics that no longer work.
