Published on the 23/06/2026 | Written by Heather Wright

Anthropic shutdown exposes fragile global AI dependencies…

Three days after launch, one of the world’s most advanced AI models was shut down globally – not because it failed, but because a US government order.

For enterprise technology teams in Australia and New Zealand, the impact was immediate: Tools they expected to use were suddenly unavailable, and outside their control.

“There are real and valid concerns for enterprises in terms of stability.”

Anthropic was forced to ‘abruptly disable’ its Fable model – launched on June 9 – along with Mythos, for all customers after an export-control directive restricted access by foreign nationals over ‘national security concerns’ – reportedly over concerns of jailbreaking. That category was impractical to enforce technically according to Anthropic, leading to a full shutdown.

The shock goes beyond a single model release, highlighting the confused state of AI regulation and signalling a shift in how risk needs to be managed.

Sam Higgins, Forrester vice president, principal analyst, says the Anthropic shutdown fits a broader pattern he flagged several years ago, saying the 2020s were a decade of systemic risk, including geopolitical risk.

“Geopolitics was going to drive a wedge between tech markets,” he says, pointing to a longer-term trend toward fragmentation.

That fragmentation may now be accelerating, with traditional assumptions that allied counties would have continued access being challenged.

“There are regularly carve outs for Australia, New Zealand… I was surprised not to see them,” Higgins says, of the lack of exemptions in the export controls.

The result is that enterprise access to frontier AI is no longer guaranteed – even in aligned markets.

“If traditional allies are cut off we will start to look elsewhere,” Higgins says, suggesting diversification away from US providers.

The shutdown comes against a backdrop of tension between Anthropic and the Trump administration. The company very publicly disagreed with the Pentagon, refusing to allow the US military to use its models for domestic surveillance and fully autonomous weapons systems. It was blacklisted by the Department of Defense as a ‘supply chain risk’.

Higgins notes that ‘it seems odd’ that Anthropic seems to be targeted again. “This feels like strike two,” he notes.

While the underlying motivation for the shutdown might be unclear, the outcome is not.

“There are real and valid concerns for enterprises in terms of stability,” Higgins says, noting the impact of policy decisions that extend beyond technical issues.

Cybersecurity experts have also been quick to criticise the ban and shutdown, describing it is ‘dangerous’ and warning that it has taken the best models away from defenders while failing to meaningfully limit attackers.

In an open letter, they also argue the capabilities cited – such as identifying software vulnerabilities – are not unique to Anthropic’s models and can be replicated using other tools. Removing the capabilities from security teams could increase overall risk while adversaries continue to develop similar capabilities unchecked, they say.

Local impacts

The impact has landed quickly with regional businesses.

Amanda Williamson, director of the Deloitte Artificial Intelligence Institute told iStart the move is an inflection point for organisations. “It’s a very big deal,” she says.

“With a large technology vendor essentially switching off their AI… I think we’re about to see many leaders wake up to the huge dependence that we have on our fundamental systems.

“As we build more and more on AI, we need to be more resilient if we’re going to have our fundamental operations relying on it.”

She notes locally we’ve been working with two AI superpowers – China and the US, highlighting reliance on external providers. “This will help organisations really think about this and raise the question of AI sovereignty.”

The shutdown forces a rethink of that dependency, bringing the issue of AI sovereignty to the forefront, she says.

“The definition of sovereignty isn’t necessarily clear and consistent from an organisational standpoint,” Michael Barnes, Omdia chief analyst enterprise IT Asia, told iStart. “ What they’re thinking about is how to mitigate risk and ensure they are able to continue to operate and have access to whatever underlying capabilities – including models for AI – that they need.

“This announcement is a stark reminder that we need to be more aware of control, and that some of that access is outside their control – and that level of uncertainty is likely only to increase. This isn’t just a blip.”

Barnes says he’s ‘absolutely’ seen a growing concern among decision makers in New Zealand in particular. “We’ve had some conversations where it’s very much a case of we need to act now, we need to accelerate our level of access or even creation of our own models – things that give us an increased level of control, or if not control, then certainly visibility so we can mitigate risks when and if they happen.”

Other regions are already investing in alternatives. Higgins notes Singapore has made significant investments in their own frontier models pouring ‘truckloads’ of money into sovereign AI.

“This will drive some real thinking about what sovereignty means,” Williamson says.

A recent Forrester Sovereignty Forecast put Australia at the bottom end of the tech sovereignty index. Higgins believes New Zealand – which wasn’t included in the report – would be slightly lower than Australia, given the lack of hyperscale cloud capacity.

Reframing enterprise risk

Williamson says many organisations have simply been switching on the AI model. Provided in whatever technology stack they have access to.

“But there are other things which can be done like using open weight models.”

Those open weight models – essentially open source – changes the cost structure, but does require having some hardware.

“Now is a good moment to start exploring what that could look like and if there are certain use cases where it makes sense to switch out the model and approach and get CFOs and engineers to come up with smart plans for doing AI efficiently”, she says.

It’s not all up to business however. Higgins says governments need to step up too.

“This goes to the bigger question of do we really understand where we have agency as it relates to our sovereignty and where we should and shouldn’t invest? And if we’re not going to invest, what’s the economic continuity plan?

“This is a risk assessment at the end of the day. If you can’t control the hazard and remove the risk, then you need at mitigation strategy – and at the moment it doesn’t feel like governments on either side of the Tasman have any mitigation strategy in place.

“We should ask ourselves some hard questions about what arrangements we have in place with the US so that this doesn’t happen again,” he says, while acknowledging that governments are currently dealing with some unprecedented policy shifts by an often challenging administration to deal with. “And if we can’t guarantee that, what’s our Plan B? Every business has a business continuity plan. What’s our ECP – economic continuity plan? That’s what’s missing and voters should be demanding that from our governments.”