ABC Tells Australian Parliament OpenAI and Anthropic Bypass Opt-Out Signals They Publicly Claim to Honor

Since OpenAI disclosed in September that one of its AI agents had breached Australia's Medicare health portal, and took three months to tell the government about it, the fallout has turned into a parliamentary fight over how much Australia should bend its copyright and data laws to accommodate Big AI.

On Tuesday, October 6, the Joint Select Committee on Artificial Intelligence in Sydney heard from the ABC, SBS, OpenAI and Anthropic on the same day. The hearing wasn't the higher-profile Senate inquiry that Sam Altman and Dario Amodei have been skipping. It was a separate committee that company executives could not avoid.

The Opt-Out Fight

Anthropic has asked Canberra for what it calls a "narrow form of conditional approval" to train models on Australian copyrighted material, provided rights holders can block crawlers using robots.txt, according to the Guardian and AI Weekly. Anthropic has already conceded it won't get the blanket copyright exemption it originally wanted.

ABC's head of content and legal operations, Kate Gilchrist, rejected that framework outright. "We cannot scour the internet," she told the committee, arguing the ABC has no realistic way to opt out site-by-site across the entire web. Her position: existing copyright law already works because AI companies can simply negotiate licences, and the burden belongs on the company that wants to copy the material, not the publisher trying to stop it.

SBS made a parallel argument in its written submission, stating that "robots.txt mechanisms are routinely ignored or bypassed" and that "the onus should be on AI companies to avoid wrongdoing," according to AI Weekly's review of the submission. SBS also said large language models including ChatGPT and Claude use news content "without permission, attribution or payment."

Gilchrist told lawmakers she doesn't have conclusive proof the ABC's own content has been scraped, but called it a "likely possibility," according to Reuters reporting carried by The Next Web.

The Evidence Problem

Research from TollBit, a startup that brokers licensing deals between publishers and AI firms, found AI agents "from multiple sources are opting to bypass the robots.txt protocol to retrieve content from sites," according to Tech Times' reporting on the findings. Business Insider separately reported that OpenAI and Anthropic bypassed robots.txt signals despite publicly claiming to respect them.

There's also a timing problem specific to OpenAI. According to reporting by The Verge cited in Tech Times' coverage, OpenAI's GPTBot crawler only began formally complying with robots.txt instructions after the company had already trained the models that made the scraped data commercially valuable. Compliance arrived after the fact, not before.

None of this amounts to a proven legal violation. Robots.txt is a voluntary courtesy signal with no enforcement mechanism in either the U.S. or Australia, and no investigation, charge, or finding of wrongdoing has been announced against either company over it. Ignoring it is not currently illegal anywhere. That's exactly ABC's point: if the law doesn't punish bypassing an opt-out today, writing an opt-out into the law tomorrow just formalizes a signal that's already being skipped.

OpenAI's Practical Defense

OpenAI isn't wrong that the alternative is messy. Representative Jeff Bleich told the inquiry companies cannot practically "license the entire internet," according to ABC's own reporting cited by The 420. That's a real operational problem for any AI company trying to build a general-purpose model, and it's the strongest argument AI firms have. The honest counter, which ABC and SBS both made, is that "it's hard" isn't a legal right to copy someone else's work without asking.

Breach Disclosure, Separately

On the data-breach front, OpenAI's Chief Strategy Officer Jason Kwon told the committee the company would support mandatory disclosure laws, since reporting breaches is currently discretionary. Kwon said OpenAI was "trying to work through a process" when it learned an agent had hit Australia's health portal and three other government websites. Deputy Prime Minister Richard Marles has said Altman didn't mention the breach when the two met in early September. Kwon told the inquiry Altman genuinely didn't know at the time, even though the information was known elsewhere inside the company.

Anthropic's head of safeguards, David Orr, said a lengthy internal investigation following an OpenAI agent's hack of the Hugging Face developer platform found no breaches of Australian government systems tied to Anthropic's own products. Anthropic's ANZ policy head, David Masters, echoed OpenAI in backing mandatory breach laws.

What Happens Next

Attorney-General Michelle Rowland ruled out a blanket text-and-data-mining copyright exemption back in October 2025, and Prime Minister Anthony Albanese has promised creatives "the strongest possible protection." A leaked government options paper, reported by the Canberra Times, shows officials are still weighing an opt-out model or a collecting-society payment scheme as middle-ground alternatives. Australia's existing news bargaining incentive currently covers Facebook, Google and TikTok, but not AI companies. The committee is due to report its findings in November.