The Sword Cuts Both Ways: Professor Toby Walsh on AI, Mythos and the New Normal in Cyber
Cyber VoicesShow Notes
On this episode of Cyber Voices, host David Savva-Willett is at Canberra CyberConnect 2026, AISA's inaugural event in the nation's capital, for a wide-ranging conversation with Professor Toby Walsh, one of the world's most influential voices in artificial intelligence.
Toby is a Professor of AI at UNSW Sydney and Chief Scientist of UNSW AI. He has advised the United Nations and heads of state on the limits we need to place on AI, and his outspoken stance on the military uses of the technology famously earned him an indefinite ban from Russia.
In this conversation, Toby and David dig into what AI really means for cyber defenders right now. They discuss Anthropic's Mythos and the wave of decades-old zero-day vulnerabilities now being uncovered, why this is the new normal rather than a one-off event, and how AI has democratised offensive capability so that sophisticated attacks no longer require deep technical expertise.
They also explore the questions that matter most for security leaders: whether defenders are really losing the AI arms race, why dwell time has collapsed from 200 days to a smash-and-grab measured in hours, the rise of shadow AI arriving both top down and bottom up, the sovereignty risk when powerful tools are released only to a select few, and the lessons from the Canvas breach where attackers did not hack the front door, they simply logged in.
Toby also lifts the lid on the ideas behind his latest book, The Shortest History of AI: Six Ideas Are All You Need to Know, including why AI is a 70-year overnight success and why the human brain, running on the power of a dim light bulb, still puts our most advanced machines to shame.
Whether you are a CISO being asked to govern AI while still learning it yourself, or simply trying to separate the signal from the hype, this is a clear-eyed and occasionally very funny look at where AI and cyber security collide.
Topics covered: Why AI is a double-edged sword for cyber, threat and defence at once Anthropic's Mythos and the discovery of zero-day flaws nearly 30 years old How AI has lowered the barrier to entry for sophisticated attacks Whether defenders are losing the AI arms race Dwell time collapsing from 200 days to under two hours Shadow AI, and how security leaders can actually govern it Sovereignty risk and the case for stronger regulation The Canvas breach and the era of just logging in Six big ideas from The Shortest History of AI
Cyber Voices is the official podcast of the Australian Information Security Association (AISA).
More from Cyber Voices
Recorded on the CyberCon 2026 expo floor in Melbourne, this episode sees host David Savva-Willett sit down with Keren Elazari, the friendly hacker, the day after her keynote closed out day one of the conference. Keren is the founder of BSidesTLV and Leading Cyber Ladies, a senior researcher at Tel Aviv University, and the speaker behind one of TED's most watched talks on why the world needs hackers.
Recorded on day three of CyberCon 2026 in Melbourne, the first episode of the Cyber Voices CyberCon series brings AISA Chair Michael Burchell back into the booth with host David Savva-Willett. Close to 5,800 delegates came through the doors this year, a record for the conference, and the pair reflect on what made it work: a program with a record number of speakers presenting for the first time, a deliberate focus on cyber while a global trend pulls conferences towards AI, and a community that turns up to share what it has learned.
There was a pen on the table at a workshop. Placed next to an unlocked phone, it opened the browser and loaded the facilitator’s website, with nobody tapping anything and nobody agreeing to anything.
Information stealing malware is not new and not sophisticated. That is precisely what worries Jesse Hoppo, who leads the open source intelligence and cyber crime team within Telstra's Threat Research and Intelligence Group, and who spends a good deal of his time reading what these things send home.
Governance, risk and compliance generates a great deal of manual work, and not all of it is work anyone would miss. Ian Yip's argument is that this is exactly where AI earns its place in a security team, provided nobody mistakes it for judgment.