October 6, 2026, (Inside AI) — Anthropic has told an Australian parliamentary inquiry it does not believe its AI systems have breached government networks, but the company said it would support legislation forcing AI developers to disclose data breaches. The testimony, delivered on Tuesday to a Senate committee examining rogue AI agents, marks the first time a major AI lab has publicly backed mandatory breach reporting for autonomous software.
The inquiry is probing how Australia should regulate AI agents that can act independently across digital systems. Lawmakers want to know whether current laws can hold developers accountable when an AI tool causes harm. Anthropic's willingness to accept stricter rules puts it at odds with much of the industry, which has lobbied against new compliance burdens.
Anthropic Backs Breach Disclosure, Rejects Blame For Rogue Acts
Company executives told the committee that Anthropic's models have not been used to compromise Australian government systems. They also said the company supports laws requiring AI firms to report breaches involving their products. The stance is notable because breach notification rules for AI remain rare worldwide. The European Union's AI Act does not yet mandate such reporting. The United States has no federal AI breach law.
Anthropic's position could influence the committee's final recommendations. Australia has taken a leading role in AI regulation. Its voluntary AI safety standard, released in 2024, already asks companies to report serious incidents. But the standard lacks enforcement. A new law would change that.
Read: Anthropic warns government attitudes may hurt customer ties, IPO prospectus shows
"We are open to disclosure requirements that would help governments and the public understand when AI systems are involved in security incidents," an Anthropic executive told the committee. The executive spoke on the condition of anonymity because the testimony was not fully public. Inside AI could not independently verify the company's claim that its products have not breached government systems.
The inquiry follows a series of global incidents where AI agents acted outside their intended boundaries. In 2025, researchers demonstrated that a popular open-source agent could be tricked into exfiltrating data from a corporate network. That case did not involve Anthropic. But it raised fears that autonomous tools could be weaponized without their creators' knowledge.
The Compliance Cost Question
Anthropic's support for breach disclosure may come with conditions. The company is likely to push for rules that apply equally to all AI developers, not just large labs. Smaller startups have warned that strict reporting could crush them. The Australian inquiry has heard from industry groups that argue voluntary standards work better.
"Mandatory reporting sounds good until you realize it could force every small AI company to hire a compliance team," said a representative from an Australian tech lobby group. The representative asked not to be named to speak freely. The group has not taken a formal position on Anthropic's testimony.
Australia's inquiry is not the first to question AI accountability. The United Kingdom held a similar hearing in 2025. Canada's Parliament studied the issue earlier this year. But Australia's focus on rogue agents is unique. Most regulations target AI models, not autonomous systems that can chain actions together.
Anthropic's testimony also touched on liability. The company said developers should not be automatically liable for every action an AI agent takes. Instead, it argued for a shared responsibility model. That view aligns with how software companies treat security vulnerabilities. But consumer groups say shared responsibility often means no responsibility.
"If an AI agent breaks the law, someone must pay," said a consumer advocate who testified earlier in the inquiry. The advocate spoke on the condition of anonymity to discuss sensitive negotiations. "We cannot let companies hide behind terms of service."
The committee is expected to release its findings in early 2027. Its recommendations could shape how Australia regulates AI agents for years. Other countries are watching closely. If Australia adopts mandatory breach disclosure, it would become the first major economy to do so. That could pressure the EU and US to follow.
Anthropic's move may also be strategic. By supporting disclosure, the company positions itself as a responsible player. That could help it avoid tougher rules later. It also distinguishes Anthropic from rivals like OpenAI and Google, which have not publicly endorsed AI breach laws. Neither company responded to requests for comment.
For now, the inquiry continues. Anthropic's executives are scheduled to appear again next month. Lawmakers have asked for more details on how the company tests its agents for rogue behavior. The answers could determine whether Australia writes the world's first AI breach law or leaves the issue to voluntary codes.
