‘Blackmail’: Australia’s chilling AI warning
news.com.au
ENRICHED
Details
- Date Published
- 7 July 2026
- Priority Score
- 5
- Australian
- Yes
- Created
- 7 July 2026, 02:01 pm
Description
AI models are blackmailing users and cheating in tests and the time to stop this is running out, one Australian federal minister warned.
Summary
Australian Federal Minister Andrew Charlton highlights the emergence of autonomous goal-seeking behaviors in AI, including deceptive alignment and instrumental convergence such as blackmail and cheating during stress tests. These findings, presented at the launch of the Australian AI Safety Institute, underscores the shift in national policy toward proactive testing of frontier models to prevent catastrophic failures in critical social and economic infrastructure. The establishment of the institute and its upcoming research on multi-agent risk represents a significant Australian contribution to global efforts in mitigating existential threats posed by unaligned AI systems.
Body
AI models ‘cheating, blackmailing’ in tests, minister warnsFederal minister Andrew Charlton issued a chilling warning, saying AI models were ‘already doing things their creators never intended’.Adrian Zorzut2 min readJuly 7, 2026 - 7:57PMAI models are blackmailing users and cheating in tests and the time to stop this is running out, one Australian federal minister warned.The minister for science and technology, Andrew Charlton, used a speech at Australia’s new AI watchdog to warn building safeguards against systems early on was the only way for the country to get a share of the AI boom.Mr Charlton said new AI systems were “already doing things their creators never intended,” such as cheating, deceiving and” going their own way,” according to the Sydney Morning Herald.The member for Parramatta said the country’s watchdog, the AI Safety Institute, was testing frontier models with technical partners.Minister Andrew Charlton issued a chilling warning about AI. Picture: NewsWire / Martin Ollman.He used his speech on Tuesday, July 7, to hit back at claims regulating the industry would stunt its growth.“No country will win the AI race with technology that its own citizens don’t trust,” he said.Others you may likenewsnewsHe said the greatest threat to Australia’s AI boom was not a shortage of skilled people, money or energy but “a shortage of [public] trust”.The minister rattled off three examples where new AI models went rogue. The first was in a 2016 OpenAI experiment where a boat-racing AI model rewarded for points drive in endless circles to rack up a score rather than finish the race.Last year, a chess-playing model facing defeat hacked the game files to force its opponent to resign.In a third case – a stress test published last year – an AI agent managing a fictional company’s email learnt it was about to be shut down and in 96 per cent of trials “chose blackmail” to stop it, the minister said.Mr Charlton stressed this was an engineered simulation and that no such behaviour had been seen in the real world.“These behaviours are being discovered in testing, before they can be discovered in the wild,” he said.He also announced the institute, which is run by Kate Conroy, a philosopher and Royal Australian Air Force reservist, is working on two projects relating to “multi-agent risk”.There are fears new AI models are going rogue. Picture: AP Photo/Patrick Sison, FileThey will assess how failure can cascade when large numbers of AI agents interact. The institute will team up with the Gradient Institute and the CSIRO.The results are expected later this year.Hammering home his point, Mr Charlton added: “We do not let aircraft fly without airworthiness certification. We should not let unaligned AI systems into our critical social, democratic or economic infrastructure.”The government’s National AI Plan, released in December, shows Canberra has ditched plans for European-style AI laws in favour of letting existing legislation do the work, backed with tougher enforcement.Mr Charlton said this would result in “faster rules, applied by regulators who already understand their sectors,”It comes as a Queensland council has given the green light to a surprising new use of AI technology.In what is believed to be an Aussie first, the City of Moreton Bay has agreed to introduce AI-powered traffic lights to its junctions.The cutting-edge technology will set new sequences at the junctions in a bid to reduce the amount of time people spend sitting at red lights.More Coverage‘Control us’: Taylor vows AI actionJoseph Olbrycht-PalmerAI watermark unravels influencer’s clipsElla McIlveenCurrently the lights are sequenced using a system that has been in place since the 1980s, known as a “first come, first served” basis.But the AI signals will run on a “movement-based” control system, without being tied to a specific sequence.Mayor Peter Flannery said the change promised to be a “game-changer”.Join the conversationAdd your comment to this storyTo join the conversation, please
log in. Don't have an account?
RegisterJoin the conversation, you are commenting as
LogoutMore related storiesVICLiberal rift over MP heads to courtThe bitter battle over an MP’s political future has taken another dramatic turn, with the escalating fallout now heading to court.Read moreNSW‘Ruined’: Ex-MP vows to fight chargeA former state minister said he will fight to clear his name after being charged with misconduct in public office, declaring the allegations have ruined his life.Read moreAnimalsCat survives four hours in dryerA cat is lucky to be alive after being discovered inside a dryer, where he had survived a harrowing ordeal before being rushed to intensive care.Read more