Chairs of key federal parliamentary committees have warned that an onslaught of AI-generated material is introducing incorrect or even “hallucinated” information into policymaking, cautioning MPs to be vigilant about the veracity of material.
Analysis by Guardian Australia this week found inquiries are being flooded with content created by AI, which is inventing studies and attributing nonexistent research to real academics and authors. In some cases, committee reports have even cited submissions in which the majority of sources appear to be AI-generated hallucinations – which occurs when large language models (LLMs) produce or refer to content that looks real, but doesn’t actually exist.
The Liberal senator Paul Scarr said the reporting showed there were clear dangers in an increase in the number of submissions that included hallucinated content.
“The biggest risk is that decision-makers, whether or not they’re in parliament or in business or anywhere else, make decisions based on incorrect information,” he said. “And it increases with the standing of the person or entity that’s making the submission and we all need to be alive to that.
“We’ve been put on notice. We really need to check and go the extra mile in terms of checking footnotes and sources and claims which are made in submissions.”
Scarr was on a committee in 2023 that received a submission including fake allegations of wrongdoing by KPMG.
Guardian Australia reported at the time that the consultancy firm complained to the committee, which one academic took responsibility for.
Scarr said that as committees must consider all submissions, “you would hope that it’s evened out somewhat if there are a minority of submissions that have got these errors, but the danger’s there”.
Louise Miller-Frost, a Labor MP and chair of the social policy and legal affairs committee, which received a submission with hallucinated references, said committees were becoming more aware of the issue, but that submissions created using AI should not be discarded.
She said the findings of the submission with hallucinated content was not different to other evidence given to the inquiry.
“I think if they put in something unusual or really salient then you’d have to go down the rabbit hole and go, ‘OK I need to find out more information on that particular submission and then go into the references,’ and then it would be obvious that their references don’t exist,” she said.
“I think committees … and committee secretariats are a lot more aware of it now. Some of these committees are dealing with massive quantities of data coming in, which makes it a little bit difficult to check every single reference.”
Miller-Frost also noted that AI could be helpful for members of the public who might otherwise have difficulty making submissions.
“Some of [the submissions] are from agencies and academics and government departments and they’ve got lots of resources. But we also get, particularly in the social services area, some people who are individuals in the community, and they may not have the time and the capacity and availability to write what they want to write.”
The Senate president, Sue Lines, said MPs involved in parliamentary inquiries always tested information presented to them, but said individuals behind submissions were responsible for the veracity of the content.
“No doubt committee practices will continue to evolve as the widespread adoption of AI continues, including responding to the likely increase in submissions through the availability of AI tools – which is beneficial in terms of participation but also has an impact on the task of digesting submissions and on resources,” she said.
“Committees also set a clear expectation that where an inquiry participant becomes aware that any evidence they have given is misleading they have an obligation to correct it.”
Senate advice for inquiry submissions warns that use of AI can present risks to the quality of information and that accuracy is the responsibility of the submitter.
