Independent senator Fatima Payman has questioned senior public servants about citation errors in a $3.48 million age-assurance technology trial report, prompting an apology from department secretary Jim Betts at Senate estimates.

The report, prepared under a technology trial involving the Age Check Certification Scheme, attracted scrutiny after errors were identified in references and links. The government commissioned the work as policymakers considered how online services might verify age and manage access restrictions for younger users.

Payman pressed officials on explanations previously given to a parliamentary committee. Earlier evidence had referred to a contractor’s account that links had broken after the report was produced. Officials said they had relied on that explanation and had not independently checked every historical link.

The department subsequently accepted that multiple citations required correction and published changes on its website. At estimates, officials rejected the suggestion that they had deliberately misled the earlier committee, saying they had taken the contractor’s statements in good faith.

Betts apologised to the committee and accepted that the reporting was sloppy. He nevertheless argued that the errors did not undermine the core findings of the trial. That is a substantive position which should be presented as the department’s assessment, rather than as an independently established conclusion.

Payman’s challenge went beyond proofreading. In a government-commissioned technology report, the reliability of referenced evidence affects the ability of policymakers and the public to test its claims. Citations help readers trace assertions about technical capabilities, privacy, implementation and international approaches.

Missing or incorrect links do not necessarily falsify a report’s methodology or conclusions. But a pattern of citation failures can weaken confidence in document review and quality assurance. Distinguishing between a broken hyperlink, a wrong source and an unsupported substantive claim is critical.

The reported cost of the trial, $3.48 million, is another reason parliamentarians are entitled to ask how it was checked. Public procurement is judged not only on whether a report was delivered but whether the evidence can withstand independent scrutiny and whether errors are acknowledged and corrected transparently.

Age assurance is an especially sensitive policy area because of its potential effect on children, parents, technology providers and adults who may be asked to demonstrate eligibility. Technical findings may shape identity processes and data-protection choices. That makes clarity about the strength and limitations of the evidence important.

The hearing did not prove intentional deception by the department, its officials or the contractor. Equally, Betts’s apology did not end the legitimate question of whether existing review procedures were sufficient. Payman’s questioning brought both concerns into the public record.

The next test is whether the published corrections clearly identify the affected citations and whether policymakers explain which findings remain supported. For the public, the issue is how to distinguish a correctable documentation problem from a failure serious enough to change policy conclusions.

The immediate dispute concerns citations, but the broader context is a government-commissioned evaluation of age-assurance technology. These systems attempt to determine whether a person is above a specified age without necessarily identifying them by name. Different technologies make different trade-offs between convenience, accuracy, privacy and accessibility. Because the resulting research may inform policy affecting millions of users, source accuracy is not merely a cosmetic matter for the document’s editors.

The trial was run by the UK-based Age Check Certification Scheme, or ACCS. Guardian Australia previously reported multiple citation problems in the commissioned report and that the company had acknowledged using ChatGPT in editing. That latter detail should not be mistaken for proof that an AI system fabricated a particular reference. An editorial tool can be used in many ways; responsibility for checking the final product rests with the humans and organisations responsible for it. The specific errors have to be assessed in their own right.

Officials told a committee they had understood some reference links to have broken after the report was prepared, an explanation said to have come from the contractor. The department subsequently accepted errors and published a correction. The dispute therefore includes the quality of the original document, the speed of its correction and the quality of the explanation delivered to Parliament. Each is a separate accountability issue, even if none proves deliberate dishonesty.

In the estimates hearing, Payman asked officials about what had been told to the earlier inquiry. The department said it relied on the contractor’s account in good faith and had not independently checked all the historical links, a task officials described as difficult. Betts apologised, called the reporting sloppy and argued that the affected citations were technical footnotes rather than matters that changed the study’s principal conclusions. His view of their significance is a stated departmental position, not a substitute for an independent review of the corrected report.

The distinction between faulty references and faulty findings is essential. A hyperlink can stop working because a website changes its structure even when the referenced study exists and says exactly what an author claimed. A citation can also be wrong because it points to a source that never supported the claim, or to a source that cannot be located. These problems range from fixable formatting to genuinely missing evidence. Without examining the corrected footnotes, a reader cannot responsibly assume that every problem falls into one category.

For a taxpayer-funded project, the procurement and oversight questions are substantial. What checking did officials expect from the contractor? Was the report reviewed independently before publication? Were corrections logged in a way that allows readers to see exactly what changed? And were the same quality standards applied to technical chapters and to the executive conclusions? A clear correction record would help distinguish normal editorial repair from failures of substantive review.

Age estimation technologies can use facial analysis, third-party identity verification, digital credentials or other approaches. Each approach raises questions beyond whether it can guess or confirm a person’s age. The policy implications include the handling of personal data, the risk of excluding users who lack standard documents, error rates across demographic groups and whether alternatives are offered to people who cannot use a particular system. Those are examples of matters an evidence-based report should address, rather than findings that this report necessarily got wrong.

Accuracy also has several dimensions. A tool might be very good at identifying adults substantially older than the threshold while performing less well for people close to it. False acceptance and false rejection affect different groups in different ways. The significance of the technology findings depends partly on the testing conditions and the definition of successful verification. That is why a technically dense report needs citations that allow specialists to inspect underlying studies instead of relying only on broad assurances.

The government’s policy interests extend beyond the trial’s publication. Lawmakers are considering how online services should prevent under-age access while minimising unnecessary collection of identity information. Errors in supporting documents can erode confidence even where the underlying policy goal is widely supported. Transparent correction is therefore part of building trust in future decisions, especially when services may eventually be required to adopt new systems.

Payman’s scrutiny does not amount to a finding that the whole $3.48 million was wasted. The cost covered a broader project, and the public discussion of six citations does not establish that every part of the trial was defective. Equally, Betts’s assessment that the core conclusions survived the corrections does not eliminate a need to explain why the errors occurred or how further problems will be prevented.

Parliamentary accountability is strongest when the questions and the answers are equally visible. Officials should be able to correct errors without those corrections being treated automatically as admissions of intentional misconduct. Senators should also be able to test whether a claimed clerical mistake may have implications for reliability. The Payman–Betts exchange demonstrated precisely why those two principles must operate together.

The next useful step for the public record would be an accessible list of corrected citations alongside the claims they originally supported, with an explanation of any substantive consequences. That would allow journalists, academics and policymakers to evaluate the evidence without guessing. A technical report is ultimately valuable only to the extent that its findings can be checked. The hearing turned that basic standard into a clear issue of public accountability.

An important consideration is how technical references are independently checked when a contractor delivers research to government. Reviewers may be able to confirm that a citation exists without deciding whether the referenced study supports the proposition for which it was used. More rigorous review asks both questions. A report can pass a superficial link check yet still rely on evidence that is outdated, limited or applied outside its original context.

A transparent correction record would preserve the document’s revision history. Readers need to know which version they are looking at, when an error was reported and what specific text or references were changed. If the department regards a correction as having no effect on conclusions, explaining the reasons can help build confidence. If a change alters a major assertion, that should also be plainly acknowledged rather than buried in a footnote.

For Payman, the issue is whether Parliament received reliable explanations about the report. For Betts, the response is that officials acted on information supplied in good faith and the substantive findings remained intact. Neither position justifies claiming that intentional deception was proved. The stronger public-interest conclusion is that evidence-handling and quality assurance require transparent standards, particularly where future technology regulation may rely on the results.