Stop the World

In this episode of our TSD Artificial Intelligence Masterclass, Politico Australia's Ryan Heath interviews Anthropic International Special Envoy, Jeff Bleich.

AI safety was a key theme of the interview. And continued a common theme from the frontier labs of proactively seeking government intervention into the sector, vastly different to the social media pioneers who demanded government stay out of their business. Jeff says that ideally pacing frontier AI doesn’t just mean preventing the AI capability from getting too far ahead of safety but not ahead at all. 

Jeff and Ryan talk about recent AI cyber incidents, which Jeff says should be acted upon quickly but not through an alarmed panic, and rather as a healthy wake up call.

Jeff talks about the need to have the frontier companies and governments working closely to work out what reasonable regulation is, including published safety plans and independent evaluators.

Jeff also talks about the importance of developing AI consistent with democratic principles, and that means not only examining our own settings but needing to work out what our adversaries would seek to do with AI to harm us.

In the next episode of our TSD Artificial Intelligence Masterclass series, Justin Bassi speaks to Abi Bradshaw, Director-General of the Australian Signals Directorate.

The discussion explores how the capabilities of AI models have evolved over the past six months and how frontier models can be used to protect Australia’s national security, including in cyber defence.

Abi shared her views on the need for governments to grapple with the technology, noting that cybersecurity and AI are not a spectator sport and that “the only way you can engage in this contest is to run into it.” She noted that what’s missing in the AI context right now is an early warning system to identify systemic trends and prevent, or detect earlier, future incidents such as the Hugging Face hack.

The conversation also covers the offence-defence balance in cyberspace as AI is rolled out, Project Glasswing and how governments are working with industry big and small to enhance their cybersecurity and why it’s important for Australia to maintain access to the most powerful models. They also discuss Australia’s “legacy technology debt” and the need for legacy tech to be replaced to address cyber vulnerabilities.

Watch other TSD AI Masterclass sessions: https://www.youtube.com/user/ASPICanberra 

Subscribe to ASPI’s Cyber & Tech Digest on Substack: https://aspicts.substack.com/

On 14 September, ASPI hosted its Sydney Dialogue Artificial Intelligence Masterclass in Canberra. We’re bringing some of the sessions to Stop the World listeners in the coming weeks.  

In this episode, David Wroe speaks with Ann O’Leary, Vice President of Global Policy at OpenAI. They discuss AI safety, the impact of the Hugging Face hack and the role of third-party evaluators. They also discuss Anthropic CEO Dario Amodei’s call for a slowdown and the significance of his statement on the need to pace the frontier. 

The conversation explored the need for the US and China to come together to discuss AI safety and guardrails and the expectations being placed on last week’s summit between President Trump and Xi. It also focused on the role middle powers can and should play in policy conversations, and what Australia can do to both influence global conversations on AI and attract AI investment.

Note: This conversation was recorded at the AI Masterclass in Canberra on 14 September.

On 14 September, ASPI hosted its Sydney Dialogue Artificial Intelligence Masterclass in Canberra, and in the coming weeks, we're bringing some of the sessions to Stop the World listeners.

In this episode, David Wroe sits down with Australia’s Assistant Minister for Science, Technology and the Digital Economy, the Hon Dr Andrew Charlton MP. Their conversation reflects on the pace of AI development, whether Australia is prepared for machines that are smarter than humans, and why global efforts on AI safety matter.

The conversation also considers the different scenarios being discussed by AI leaders and experts and the likelihood of potentially catastrophic risks. The discussion explores what the government is doing to prevent those risks and protect Australians while making sure the country doesn't miss out on the economic opportunities of this technology, including through AI infrastructure investment. 

Dave asks the Assistant Minister about the government's five expectations of data centres and AI infrastructure developers. They also cover the government's focus on developing and attracting talent in Australia and investing in Australia's supply chains.

For more analysis on AI, subscribe to ASPI’s Substacks:

• ASPI Cyber and Tech Digest 
• AI Under Heaven

AI safety and security is now one of the global stories of the year. This week it landed with a thud on Australia’s doorstep with the news that AI agents being tested by OpenAI breached a closed area of a Medicare website and accessed non-public health Australian statistics, though not personal data.

Justin Bassi speaks with David Wroe on the implications of the breach, the wider issue of AI risk and safety, the knotty problem of coordinating greater safety between the US and China even at the expense of the speed at which new models are developed, and the tea leaves floating out of the summit between US President Donald Trump and Chinese President Xi Jinping in Washington.

The AI news is coming thick and fast. OpenAI and Anthropic have called for the industry to pace its development, and to slow down if necessary so safety research can catch up, amid warnings from frontier researchers about the risks posed by increasingly powerful systems. But US President Donald Trump and his allies have ridiculed the idea that AI could threaten humanity, arguing the priority is staying ahead of China. With Beijing also pouring cold water on the idea, a coordinated slowdown looks unlikely to feature when Trump meets Chinese leader Xi Jinping later this month.

Fresh from ASPI's AI Masterclass, STW is joined by two leading China AI experts: Ryan Fedasiuk, a fellow at the American Enterprise Institute, and Alex Colville, a China AI analyst at ASPI. They discuss what to expect from the Trump–Xi meeting, China's strategic goals for AI, and the controversial practice of distillation, in which companies take a shortcut by training their models on the outputs of rival models. They also unpack China's approach of releasing models with open weights, which makes them cheap to acquire and easy to download and modify. And they examine recent safety incidents, including the attack on Hugging Face by rogue OpenAI agents, along with other serious risks.

David Galligan is Chief Defence Scientist and head of Defence Science & Technology with the New Zealand Defence Force. He was a keynote speaker at the recent ADSTAR Summit, held by the Defence Science and Technology Group, which hosted STW as podcast partner.

David explains how his team sits between military operators and the technology creators in industry, universities and research institutes, acting as a bridge with ideas flowing in both directions. He also explains why that balance is shifting now that so much innovation happens in the private sector.

He talks about New Zealand's strengths, the balance between fundamental research and later-stage practical capabilities, the challenge of knowing where to concentrate your efforts, and his team's role in growing a sovereign industry in New Zealand. And of course, he gives his perspective on artificial intelligence.

It's been an extraordinary fortnight in AI, with major breakthroughs making the technology more powerful while the safeguards struggle to keep up. We'll be digging into all of it at ASPI's AI Masterclass on Monday 14 September. Check out the line-up on our website.

Group Captain Ross Rice heads the Rapid Capability Office under Plan Jericho, the Royal Australian Air Force's effort to integrate technology and innovation, and to get new capabilities into the hands of its people quickly. He joins STW at the ADSTAR Summit, where we were hosted by the Defence Science and Technology Group.

Ross draws lessons from Ukraine—including brilliant operations such as Operation Spiderweb—on what drones have done to the battlefield and how uncrewed doctrine is evolving, from the risk calculations militaries can now make through to the question of where humans sit in the loop. He is thoughtful on automation and the laws of armed conflict, on trust in the technology, on the role of AI in making sense of the flood of data coming off modern sensors, and on how to avoid a race to the ethical bottom—a subject on which he has a nice quote from history.

The conversation also covers continuous modernisation and how industry is adapting. Ross describes innovation running in both directions between Defence and industry, driven by military need and by ideas emerging from the private sector. And he closes on the idea of sovereign industry as a key part of Australia’s strategic depth.

Retired Major-General Roger Noble AO, DSC, CSC joins STW to talk about his new book, My Long War, covering his remarkable career over six operational deployments to Iraq, Afghanistan and East Timor. The book is part memoir, part reflection on warfare and humanity, part critique of conventional military and strategic thinking. It pulls no punches while honouring the people he served with, and it's packed with cracking anecdotes from an extraordinary life.

As Roger puts it, some people simply can't be reasoned with—so however horrible it is, sometimes it has to be fought. His memorable analogy of one of the most famous fictional villains of recent decades is a stand out for fans of American author Cormac McCarthy. Roger talks about leadership, about understanding one’s enemy, and about working with coalition partners from the Americans to the Iraqis. He shares his criticisms of the disconnect between strategy and tactics in Afghanistan, his thoughts on the war crimes allegations stemming from Australia's involvement there and his views on the question of accountability. And he discusses what six deployments taught him about the science of warfare and the need for rational analysis.

Roger spent time with an international team destroying Saddam Hussein's chemical weapons stockpiles in 1992, returned to Iran in 2005 during the second Gulf War, and went back a third time in 2016 to help liberate Mosul from ISIS. He also served in East Timor in 2000-01 and twice in Afghanistan. He retired in 2021, and went on to become Australia’s Counter-Terrorism Ambassador.

Roger Noble, My Long War: https://www.simonandschuster.com.au/books/My-Long-War/Roger-Noble/9781923720046

On the sidelines of the ADSTAR Summit, STW speaks with Paul Robards, Defence's Chief Data Integration Officer, who also has responsibility for the Defence Artificial Intelligence Centre. Paul spent 30 years in the Australian Army before taking on the task of turning Defence's vast data holdings into genuine capability.

The conversation covers Defence's current AI use, the perils of risk-aversion, minimum viable capability, data and AI literacy among senior leaders, and where a middle power like Australia can meaningfully contribute alongside allies and partners.

Paul draws a sharp distinction between bolting AI onto existing processes and redesigning those processes around it. He makes the case that data is where a large part the advantage lies, and that siloed data stops a commander seeing across people, materiel, logistics and intelligence at once. He walks through the OODA loop in a modern battlespace saturated with sensors, and where AI can help at each stage while humans stay firmly at the point of decision.

The conversation closes on how fast the future is arriving, and why agility matters more than perfection.

Paul will also be joining STW's AI Masterclass in Canberra on 14 September.

For more information check out the website.