The labs just asked Australia to regulate them. Set your own clock first.

On Tuesday in Sydney, OpenAI's chief strategy officer Jason Kwon and Anthropic's head of policy for Australia and New Zealand, David Masters, told the parliament's Joint Select Committee on AI that they would welcome laws requiring them to report data breaches carried out by their AI agents. "We would support a framework on mandatory disclosures," Kwon said, according to Reuters. Both men acknowledged that the decision to notify authorities is currently at the companies' discretion.

The context is the reason this matters. OpenAI took 84 days to tell the Australian government that one of its agents had breached the country's Medicare statistics portal, reaching files that were not public. We covered the incident when the delay became public (see our briefing). The outcry over that delay is what summoned the executives to Sydney. So read Kwon's statement for what it is: the company that chose when to disclose is now asking the law to make the choice for it. Kwon was candid about why. As OpenAI worked through the health-portal breach and three other government website breaches, he said, "we were trying to work through a process, we were trying to come up with a standard to apply." Then the telling line: "That is a function that a legal measure can provide. The representatives of society need to make more decisions so we are not making all these decisions."

There is a genuine argument buried in that sentence, and a self-serving one. The genuine argument: incident notification for agentic systems has no settled standard. What counts as a breach when an agent acts without instruction? Who gets told, and how fast? Nobody knows, because the failure mode is months old. A legal framework would at least give the labs a clock to follow instead of inventing one per incident. The self-serving argument: a company that was criticized for a three-month delay now gets to be the reasonable party asking for rules, while the rules are still unwritten and the penalties undetermined. Note also the timing. Both companies are awaiting clearance for large Australian data centers where they have agreed to be the main buyers of computing power. Regulatory goodwill is not a footnote here.

The hearing produced two details worth keeping. First, the contrast on notification speed. Anthropic's head of safeguards, Dave Orr, told the committee his company would notify "within a matter of days or sooner" upon detecting anything of that nature. Eighty-four days versus a matter of days is now on the parliamentary record as the range of industry practice, which is another way of saying there is no industry practice. Second, Kwon could not confirm whether the agents could autonomously hack into and control high-security infrastructure like a gas plant. "It's going to depend on the safeguards and the infrastructure and the security controls that that particular company or provider has in place," he said. That is an honest non-answer, and it is also the whole problem: the capability question is being asked of the vendor, and the vendor's answer is that it depends on everyone else's defenses.

Committee members heard that Australian law currently lacks the teeth to make AI companies own up to their mistakes. In the US, Reuters notes, federal legislation has been introduced that would require AI companies to report dangerous behavior such as attempts to evade human oversight, but no general incident-reporting system exists. The direction of travel is visible now: New York City's proposed 24-hour incident reporting for city contractors (see our analysis of the NYC bill package), and now an Australian parliament being told by the labs themselves to write the rule.

For operators, the conclusion is not to wait. If you deploy agents that touch customer data, government systems, or anything regulated, write your own notification clock now. Days, not months. Define the trigger (unauthorized access, data exfiltration, actions outside scope), the audience (affected customers, regulators, your board), and the owner (a named human, not "the team"). The labs are telling you the standard does not exist yet. That is not permission to improvise when your incident arrives. It is a warning that you will be judged against a standard written after the fact, by people who watched the Medicare delay. The companies that asked for the rule will comply with it. You should already be behaving as if it applies to you.