‘Triumph of IPO hype’
Futurist Ben Reid said the statement was eye-catching.
“But it’s worth noting the background context. This is largely a response to Anthropic’s restricted Mythos model, which has generated enormous media attention,” Reid said.
“The more sceptical among us might argue that this is as much a triumph of pre-IPO marketing strategy as genuine threat escalation.
“So treat claims of an ‘order of magnitude’ improvement with appropriate scepticism – even though the model is clearly a step up. Case in point: reports today that Mythos penetrated almost all NSA [US National Security Agency] classified systems in hours grab headlines, but whether that reflects the model’s capability or the NSA’s outdated cyber security practices is an open question.”
“The more sceptical among us might argue that this is as much a triumph of pre-IPO marketing strategy as genuine threat escalation,” Ben Reid, head of research at Memia, said.
Claude maker Anthropic and ChatGPT maker OpenAI are both lining up stock market listings at anticipated valuations near the US$1 trillion ($1.75t) mark.
Anthropic offered the likes of governments and banks a preview of its latest AI model, Mythos, citing its ability to find security vulnerabilities.
A “constrained” commercial release of Mythos, dubbed Fable, was briefly released then disabled over US “national security concerns”.
Publicity around Mythos has helped Anthropic pass OpenAI in corporate sales, by some measures.
Casing every window at once
Another expert said Five Eyes’ “months, not years” timetable, seems logical, given AI’s capability.
“We might be adding risk without meaning to, by giving AI too much freedom over systems that matter,” Deloitte New Zealand’s Dr Amanda Williamson said.
“AI’s power can be turned against us. It can hunt for weak spots in an organisation’s defences and slip through them,” Dr Amanda Williamson, director of Deloitte New Zealand’s AI Institute, said.
“Like someone looking to rob walking down a road, checking every window until one is unlocked, AI can now check every window in the city in a single night.
“The good news is that it is not too late. We still have time to act. The strongest defence against the coming waves of attacks is likely to be AI itself, in the hands of the people who protect our systems.
“AI in the hands of bad actors is a set of superpowers. Like in the movies, it helps when the good guys have superpowers too. We can fight AI with AI.”
‘Dual use gap’
Reid’s concern is that big countries, and big businesses, have more ability to fight AI with AI than the rest of us.
“The deeper issue is the ‘dual-use gap’,” he said.
“AI makes both attackers and defenders more capable, but defence doesn’t scale equally.
“Arguably, AI widens the cyber security gap between haves and have-nots – small businesses, solo software developers and small governments like ours are now operating in a materially more dangerous environment.”
In its 2025 annual report, the GCSB said it utilised its ties with global partners, and information-sharing with large organisations and corporates, to boost the scale of its response to AI and other threats.
‘Threat become obvious in recent months’
Victoria University AI expert Dr Andrew Lensen broadly agreed with Five Eyes’ timeline take.
“Generative AI, specifically Large Language Models (LLMs) are making it easier than ever to write code fast,” he said in an email.
“In recent months, it has become obvious that, because LLMs are so good at understanding code, they are also an emerging tool for finding software vulnerabilities. This can be used both for good – patching code before it is exploited – and for bad; finding new exploits before they are patched.
“It is good to see the cyber security agencies raising the alarm, as companies need to get ahead where they can in this new AI cyber arms race.”
NZ Govt’s new plan, new spending
The Government recently released a document called New Zealand’s Cyber Security Strategy 2026-2030, which describes the nature and scale of the rapidly evolving AI threat.
It calls for a whole-of-society response and says the Government chief information security officer and the GCSB consulted with government agencies over new minimum cyber security standards.
A new registry of critical infrastructure will be created.
The new strategy document was light on AI strategy. Robinson said the agentic AI threat has increased significantly even in the months since the document was created. More strategy work was being done in the area.
Tuanz recently recommended a single “national digital clearing house” that could assess vendors and new technologies like Mythos and Fable, preventing double-up work at Government agencies, and provided a valuable resource for businesses.
Robinson said that approach had its pluses and minuses. There could be a perception of playing favourites. If the Government did end up taking that approach, it would be through the Department of the Prime Minister and Cabinet.
Budget 2026 allocated $155.7 million in new operating funding for intelligence agencies over four years. That was split between $71.4m for the GCSB (which includes the National Cyber Security Centre) and $84.3m for the New Zealand Security Intelligence Service.
Combined, the two intelligence agencies’ annual appropriations rise from $405.3m in 2025/26 to $486.4m in 2026/27, an increase of about $81.1m or 20%.
Budget 2026 also saw a $153.6m boost in funding for Health New Zealand Te Whatu Ora to boost its cyber security, following recent high-profile breaches at private sector partners.
Deloitte’s Williamson said: “I suggest leaders ask two questions: first, where can we safely use AI to make our defences stronger, for example, by watching for unusual activity and finding weak spots before attackers do?
“Second, where might we be adding risk without meaning to, by giving AI too much freedom over systems that matter?”
Five Eyes’ practical tips as AI threat ramps up
“These actions are not new, but are now urgent to reduce not only technical risk, but also operational, financial and reputational exposure,” the Five Eyes advisory said.
They are:
1. Reduce your attack surface: limit unnecessary system access and external connectivity. Challenge whether systems need to be exposed at all and isolate those that do not.
2. Accelerate patching processes: AI is shortening the time between vulnerability discovery and exploitation. Delays in patching increase risk, especially for operational systems with long update cycles. Prioritise security updates accordingly to manage risks.
3. Address legacy systems: unsupported systems are easy targets. They are not just technical debt, they are strategic liabilities.
4. Review and strengthen identity and access controls: limit who can access critical systems. Enforce strong authentication and regularly review permissions.
5. Prepare for incidents before they happen: test response plans, train and prepare teams, and assume breaches will occur. Focus on fast containment and recovery.
See also the NCSC’s: Frontier AI: Managing the increasing risks from vulnerabilities
Chris Keall is an Auckland-based member of the Herald’s business team. He joined the Herald in 2018 and is the technology editor and a senior business writer.