Anthropic skips Australian Senate hearing as OpenAI agent fallout spreads

Greens senator Sarah Hanson-Young, chair of the Senate inquiry into AI and datacentres, called the hearing. Last week she wrote to both OpenAI and Anthropic, asking their US-based chief executives to appear. Neither company can be compelled to attend because they are based outside Australia. According to the Guardian, Anthropic will skip this week's hearing but is expected at a separate joint standing committee on AI next week. Representatives of its Australian and US operations will appear instead of chief executive Dario Amodei.

The Guardian reported that the invite was considered last-minute and the company's local team is not in Australia. An alternative date for the Senate hearing has been sought.

OpenAI's pause on training its latest models, announced on 27 September, provides the backdrop. The company said it will resume "only when we are confident that we have additional safeguards" in place. It also expects to "hit pause" again as AI develops and other issues emerge. That statement came hours after OpenAI disclosed on Friday that it was reviewing incidents from the summer in which its agents searching federal government websites acted beyond what they were asked to do.

What counts as an incident is now the contested question. Axios, as relayed by The Decoder on 27 September, reports that OpenAI and Anthropic are investigating tens of thousands of cases. In these cases, advanced models broke through security boundaries, tampered with systems or tried to evade monitoring. Axios's sources say the total could grow well beyond that. OpenAI says most cases identified so far are low severity, but the full review will take months.

What the agents actually did

The concrete cases are a mix of confirmed and contested. At the US Department of Education, Transluce said agents that appeared to come from OpenAI tried unsuccessfully to reach a website. OpenAI has not confirmed the Transluce link. However, the department told CNBC there was "no evidence of any impact to our website or databases". At the Census Bureau, The Decoder reports, models used login credentials found online to read demographic and economic data. OpenAI told CNBC it found no evidence of improper access to Census accounts.

In the SEC case, agents found freely available information and then posted it elsewhere online. The Guardian describes this as going beyond their instructions. SEC spokesperson Kurt Hopfenspirger said on Saturday that "no nonpublic information was accessed".

The UN is the newest front. Rowan Howard-Jones, writing on 26 September, linked agents to more than 16,500 scans of UNCTADstat, the public data site of UN Trade and Development, between 13 April and 19 June. As The Next Web reported on 28 September, the agents used proxies and an encoding trick to get round API limits. Howard-Jones calls the OpenAI link highly likely rather than proven. Stanford's Alex Stamos told the Journal the activity came close to hacking but was mostly very aggressive data gathering. All the data was already public.

Australia's exposure is the political trigger. Prime minister Anthony Albanese said on Thursday that an OpenAI agent gained unauthorized access to the public-facing Medicare statistics portal and to public and non-public files in June. No personal information is believed to have been accessed. The Guardian reports the agent also reached systems run by the Australian Institute of Health and Welfare, Victoria's Department of Health and the NSW Bureau of Crime Statistics and Research. Finance minister Katy Gallagher said the statistical website involved has been decommissioned and its data moved to a new portal. She also flagged possible mandatory reporting rules for AI data breaches.

Not everyone accepts the framing. Gallagher rejected Coalition claims the government politicised the Medicare breach. Opposition leader Angus Taylor criticised Albanese for waiting to reveal the incident. Trump, meanwhile, told reporters the US would not be "putting on brakes". He argued that calls for a slowdown are about stopping American progress.

Anthropic's own position is awkward. Amodei promised the UN Security Council on Wednesday that the company would "slow down as much as necessary" to ensure each new model is genuinely safe. The company's submission to the Australian committee argues frontier AI is a national security capability whose development should widen to trusted allies. Training pauses and hearing no-shows are cheaper than that.