AI deepfake legislation loophole could see predators escape harsher sentences

Advocates have raised concerns a loophole in proposed legislation to crack down on AI deepfakes could see offenders escape more serious sentences.

Victims' Commissioner Kate Connors welcomes the crack-down but says clashes between the new policy and existing child protection laws could see offenders plead to lesser sentences.

A parliamentary committee will report back to the government before the legislation returns to state parliament for debate.

A loophole in proposed AI deepfake laws could see predators escape with lesser convictions, Queensland's Victims' Commissioner says.

Legislation introduced to state parliament last month would criminalise the creation of non-consensual sexually explicit images or audio through artificial intelligence or other digital technology.

It would carry a maximum penalty of three years behind bars for adults, and 18 months for children.

Attorney-General Deb Frecklington told the House that the government was ensuring the state's laws kept pace with technology.

"The target of every deepfake image is a real person — increasingly, a young person whose natural vulnerabilities and often tenuous grip on self-esteem and confidence collapse with the viral spread of hyper realistic images, videos and audio that implicate them.

"The bogus content is overwhelmingly sexual in nature, with certain AI tools complicit in the peddling of this filth."

But in a parliamentary committee submission, Victims' Commissioner Kate Connors, while welcoming the crackdown, raised concerns.

She noted adults who created sexually explicit deepfakes of children would currently be charged under the existing offence of child exploitation material and face up to 20 years in prison.

Ms Connors submitted the new legislation could overlap with child exploitation laws, "which means an offender could theoretically plead to the lesser new offence".

"That would mean that the offender could receive a lesser sentence and not fall under the reportable sex offender scheme," she said.

Ms Connors recommended the proposed AI deepfake laws not apply to adults who offend against minors, to avoid the overlap.

Addressing parliament, Ms Frecklington said deepfake content "humiliates the victims who have been depicted without their consent".

But Ms Connors noted the proposed laws didn't cover 'humiliating or degrading' depictions of people and were limited to sexually explicit images and audio.

She called for the government to go further in its clampdown and expand the scope to include such content.

"In school environments, anecdotally, many cases which cause harm include content designed to humiliate and exclude and may not necessarily involve sexual content," she submitted.

"Limiting the offence to sexual content sends the message to victims of non-sexual deepfake harm that causes humiliation or degradation that their harm is not important."

The legislation differs from interstate jurisdictions such as New South Wales and Victoria, where non-consensual intimate images or audio must be created with intent.

Legal Aid Queensland submitted there needed to be changes to ensure a person was not accidentally acting illegally.

"As drafted, a person who prompts a tool for a generic image could commit an offence carrying three years imprisonment because of a resemblance they neither intended nor knew of," the submission said.

"An intent element would ensure that criminal liability targets intentionally wrongful conduct rather than accidental, unintended or legitimate behaviours."

The Queensland Sexual Assault Network accepted the rapidly evolving nature of the technology.

It called for the government to strengthen its approach with training for first responders, and education in schools and the community.

The parliamentary committee will report back to the government before the legislation returns to state parliament for debate.