Deepfakes and artificial intelligence-powered attacks are making elections harder to safeguard – and the national regulator can’t do it all alone.

Australian Electoral Commissioner Jeff Pope told a parliamentary committee on Wednesday that AI was boosting misinformation, foreign interference and exacerbating threats to political figures.

“It is a new threat that amplifies and accelerates existing threats to electoral integrity,” he said during Canberra hearings into the potentials and pitfalls of AI.

Pope’s comments came on the same day that Small Business Minister Anne Aly said she’d had more than a dozen death threats in just the past week.

“I don’t usually comment about these things, but I feel like I have to. This past week alone I’ve had 19 death threats on my social media,” Aly said in a social media post on Tuesday.

Pope floated the idea of a legal requirement to label deepfakes, but stressed the AEC’s enforcement powers were limited by its “fine-tuned” design.

It could tackle campaigning that was not properly authorised, educate the public and publish generalised debunks, but could make truth determinations only very rarely, Pope told MPs and senators.

“The more we’re asked to regulate and to make judgement, then there is a risk around that impartiality being challenged,” he said.

Pope patiently stepped through the AEC’s powers as the committee threw him bizarre scenarios involving AI. They included questions about deepfakes showing the prime minister giving “a bag of cash to someone dodgy” and biting the head off a puppy, suggested by Western Australian MP Kate Chaney.

Pope revealed the AEC had no power to contain that kind of content.

Misleading and deceptive information could be stopped only if it related to the casting of a vote, he said.

“If it was a deepfake of a senior person in government saying ‘the only way your ballot paper will count is if you complete it this way: one here, two there’ – anything else won’t count,” he said.

But he warned that banning deepfakes entirely might also run afoul of Australia’s implied freedom of political communication.

The committee had previously heard that artists and AI companies might not need any changes to copyright law. Instead, Canberra could interrogate whether there really was any need to train AI models in Australia, and let licensing markets evolve on their own, according to the Productivity Commission.

Asked about the economic upside to hosting cutting-edge “frontier models”, commission representative Stephen King replied: “That is an open question.”

His view goes against what the committee has repeatedly been told about Australia needing top-of-the-line bots and onshore training to reap the AI boom’s biggest benefits.

Experts have testified that training home-grown models may well require copyright law to change to either require tech giants to license content or give them a pass to use it at no cost.

“Copyright law would rule out the training of those models in this country, unless the party training those models licensed the relevant material,” King affirmed to the MPs and senators on Wednesday.

The government could also wait for artists and AI companies to strike deals on their own, which King said was already happening.

“Three years – let the markets develop, if they haven’t developed, if there’s still a significant problem, then then we should start considering alternative solutions,” he said.

-with AAP