AI tool ‘too dangerous to release’ could wreak havoc on businesses
Australia’s largest cybersecurity firm has issued an urgent warning about a powerful new AI tool that can find and exploit flaws in software at unprecedented speed and scale, which experts fear could trigger the next wave of major data breaches.
On Thursday, CyberCX told Australian businesses, banks and infrastructure operators they had a closing window to strengthen their defenses before the technology or copies of it fell into the hands of criminals.
The warning relates to Claude Mythos Preview, an unreleased artificial intelligence model developed by US company Anthropic that the firm considers too dangerous to be released to the public. Anthropic has restricted access to nearly 50 major technology and infrastructure partners, including Microsoft, Google, Apple, Amazon and JPMorgan Chase, under a program called Project Glasswing.
Mythos has already uncovered thousands of previously unknown software flaws, including a 27-year-old bug in an operating system used in firewalls and internet routers around the world. One test created 181 working attacks against the Firefox web browser. An older, public version managed only two.
Anthropic product leader Angela Jiang said the cyber capability emerged partly as a byproduct of the company’s broader push into coding and long-term agency missions.
“Something that is really good at coding is also very good at detecting cyberattacks, especially across a number of different surfaces, and chaining them together,” Jiang said. The company had the “privilege of working with a group of companies to help improve critical infrastructure.”
Dimitri Vedeneev, CyberCX’s security AI lead, said that what sets Mythos apart is not just its ability to find long-buried vulnerabilities, but also its capacity to chain multiple flaws together and suggest how to exploit them with a single command prompt.
“Australian organizations should not expect access to Mythos as some kind of silver bullet,” Vedeneev said. “It won’t be long before this capability, or others like it, become more widely available and potentially fall into the hands of cybercriminals.”
The mythos shook governments around the world. Bank of England governor Andrew Bailey told the BBC that central banks were examining what the technology could mean for cybercrime, while Canadian finance minister François-Philippe Champagne described Mythos as the “unknown unknown” at International Monetary Fund meetings in Washington. The Trump administration brought together leading US bank bosses to discuss the risks.
Anthropic expects rival AI companies to launch similarly powerful tools within 18 months. OpenAI has granted a select group of users access to its cyber-focused model, GPT-5.4-Cyber. Bloomberg reported last week that a small group of unauthorized users gained access to Mythos through third parties, and Anthropic confirmed that the investigation was ongoing.
Anthropic on Friday also announced the public beta launch of Claude Security, a defense product that allows enterprise customers to scan and patch their own code for vulnerabilities. He said hundreds of organizations have used the tool in research previews to find flaws that “existing tools have missed for years.” Those using the product include Accenture, Deloitte, PwC, BCG and Infosys.
An Anthropic spokesperson said the company was starting with “some of the largest US-based companies” on the basis that “if they can secure their products quickly, security will expand globally” and added that they look forward to expanding their cybersecurity partnerships.
Not everyone believes that the Mythos represents a clean break. Juraj Janosik, artificial intelligence director at cybersecurity firm ESET, said models that can detect vulnerabilities existed long before Mythos. With the right orchestration, threat actors “can already achieve Mythos-like capabilities using generally available models.”
“Many companies still lag behind in maintaining basic cyber hygiene and are often exploited using older, already publicly available vulnerabilities,” Janosik said. “While advances in AI capability are concerning, they are dwarfed by the overall lack of cyber resilience.”
Manuel Salazar, director of cyber services at Australian firm Orro, said the fundamentals had not changed. “Mythos is changing the rate at which weak fundamentals emerge,” he said. “For mature organizations, AI is a force multiplier; for less mature organizations, it can accelerate the rise but cannot bypass the basics.”
Salazar said Australian businesses should be able to access the defensive benefits of Mythos-class AI but should not have “unfettered access to the border exploitation engine”.
“Australia needs to secure access to advanced AI technologies through agencies such as the Australian Signals Directorate, Home Affairs and the National Cyber Security Coordinator,” he said. “If we don’t get involved now, we risk falling behind our U.S. counterparts.”
Wall StreetJournal On Thursday, the White House reportedly rejected an Anthropic proposal to roughly double the number of organizations with access to Mythos, citing security concerns. Management’s relationship with Anthropic had been strained by an earlier dispute over the military use of the company’s artificial intelligence and is now the subject of two lawsuits.
It is unclear whether any Australian organizations are part of Project Glasswing. Anthropic signed a deal with the Albanian government earlier this year and opened an office in Sydney, but no Australian agency has publicly confirmed access.
The warning forms the backdrop to the 2022 Optus and Medibank breaches, which exposed millions of customers’ personal information and reshaped the public’s trust in major institutions. The breaches exploited relatively traditional weaknesses, and it is feared that more advanced AI tools such as Mythos will enable attackers to find and exploit flaws in systems previously thought to be secure.
CyberCX is calling on Australian organizations to map their critical systems, segment their networks and “fight AI with AI” by using defensive AI in their security functions.
Get news and reviews on tech, gadgets and games Our technology newsletter. Sign up to receive it every Friday.
David Swan is the technology editor of The Age and The Sydney Morning Herald. He was previously technology editor of The Australian newspaper.Connect with: X or email.
