Cyber chief's AI warning amid global slowdown push

Australian politicians and AI executives have joined an international chorus of warnings around the pace of development. (ABC News: Kylie Silvester)

Abigail Bradshaw has called for Australian organisations to use AI to defend themselves and share warnings with each other about emerging threats.

Global tech executives last week warned the race to develop AI could outstrip humanity's ability to control it.

The federal assistant science minister says Australia should help shape any global slowdown effort.

Australia needs an artificial intelligence "early warning system" to help defend against the dangers of the increasingly powerful technology, according to the nation's cyber intelligence chief.

Australian Signals Directorate director-general Abigail Bradshaw also urged Australian organisations to embrace AI for its defensive capabilities, saying it was one of the crucial tools that the nation needed to protect itself.

It comes as Australian politicians and AI company executives respond to a growing international chorus of warnings the race to develop AI could outstrip humanity's ability to control it.

Speaking at the Australian Strategic Policy Institute's (ASPI) Sydney Dialogue AI Masterclass on Monday, Ms Bradshaw said intelligence and security agencies needed to work with AI companies to identify early risks.

Abigail Bradshaw says intelligence and security agencies need to work with AI companies to identify early risks. (AAP: Mick Tsikas)

Ms Bradshaw said bringing together reports of individual incidents would help identify wider patterns, adding that these risks extended beyond cybersecurity to areas such as biotechnology.

The public servant, who is tasked with Australia's cyber offensive capabilities, balanced the warning with a more optimistic assessment about AI capabilities helping those tasked with defence.

She pointed to existing cyber partnerships between government and industry, and international arrangements for quickly sharing technical information.

Australian government agencies had been granted access to major AI companies' cutting-edge models that were limited from wide release.

She confirmed that this included Anthropic's Mythos, which was temporarily restricted after the US government placed it under an export control, and OpenAI's cyber versions of their models.

Ms Bradshaw said her agency used this access to help Australian organisations fortify their systems and identify potential vulnerabilities.

"What would have taken cybersecurity experts in ASD weeks and weeks of work has been done in hours," she said.

Ms Bradshaw urged organisations to maintain basic security measures while using AI to defend their networks.

Defenders needed capabilities on par with those available to cybercriminals and state-backed attackers, she said.

In a separate panel on Monday, Australia's AI Safety Institute general manager Kate Conroy urged companies and agencies to let their teams, or their staff, experiment with advanced AI.

In July, more than a thousand employees of leading AI companies signed onto an open letter that called for the US government to step in to help pave the way for an AI development slowdown.

In the past week the high-profile resignation of Anthropic staffer Jacob Coxon and an essay published by the company's chief executive reignited a global conversation over a global slowdown effort.

Australia's assistant science minister, Andrew Charlton, told the ASPI conference that Australia should help shape the response.

Mr Charlton said governments should assume AI would keep becoming more capable, given companies' financial incentives, and despite executives discussing slowing down.

AI executives last week said the industry needed to slow development. (ABC News: Scott Preston)

OpenAI vice-president of global policy Ann O'Leary told the event the company was working with governments and AI safety bodies, such as Australia's AI Safety Institute, as well as providing models for review before release.

But, she said, the absence of wider laws slowing development meant OpenAI and other companies were employing their own voluntary co-ordination efforts to "pace" progress.

Ms O'Leary said OpenAI was eager to train its AI models in Australia, saying it was in talks with the federal government about copyright.

"We are deep in conversation with the Australian government about these questions," she said.