On 9 June 2026 Anthropic released Fable 5, which launched at #1 on the independent Artificial Analysis Intelligence Index. Three days later, a US export-control directive forced it offline for foreign nationals worldwide — Australia included.
Every hour on this clock, US citizens and companies keep building on this frontier model. Australia loses the access — and the learning curve around it. This is the case for the Australian government to negotiate access, and a workable way to do it.
Fable 5 debuts at #1 on the Artificial Analysis Intelligence Index and is adopted within days across enterprise platforms (Harvey, Databricks). The new frontier of AI capability is, briefly, open to the world.
For the first time ever, the US issues an export-control directive over access to a large language model — suspending it for any foreign national, whether inside or outside the United States, including Anthropic's own non-citizen staff.
Unable to block foreign nationals selectively in real time, Anthropic disables both models for every customer on the planet to comply. Australia loses access alongside the rest of the world. (Other Claude models are unaffected.)
Stated reason: a suspected method of "jailbreaking" Fable 5, framed as a national-security concern. Anthropic disputed the severity, and one report (TechCrunch) argued the ban "was never about a jailbreak."1,4,9
Its defining capability is long-horizon agentic work: it plans across stages, delegates to its own sub-agents, runs for days, and validates its own output. That is the difference between an assistant that helps a person and a system that does the project.
Multi-file code migrations, deep code review, security-sensitive implementations — executed autonomously over long runs.
Sustained, multi-step workflows with sub-agent delegation. The kind of work that used to need a whole team.
First model to break 90% on Anthropic's core long-running-analytics benchmark; launched #1 overall on the independent Artificial Analysis Intelligence Index.
Benchmarks are the evidence, not the point. The point: this tool can compress some skilled workflows from months into days — and right now, only people in one country can use it.3,7
The directive swept in the United States' closest allies — the UK, Canada, Australia and New Zealand — and their nationals, with no distinction from adversaries. Analysts called it a blunt geopolitical signal that Five Eyes partners should never have been caught by.4,5,6
A frontier-model advantage isn't a one-off head start you can sprint to close. Each day of exclusive use feeds four loops that make the next day's advantage larger. "We'll catch up later" assumes a gap that stands still. This one accelerates.
Work done with the best tool produces better tools, data and workflows — which do better work.
Output per worker steps up where the frontier is available, and the lead reinvests itself.
Hands-on fluency with frontier agents is experience you can't import after the fact.
Standards, tooling and capital coalesce around whoever ships first. Everyone else integrates late.
Illustrative. The curve shows the shape of a compounding-advantage dynamic, not a forecast of a specific quantity — the mechanism, not a number we have invented.
The honest answer to "what does the delay cost?" The economy-wide AI prize is large and measurable; the slice owed specifically to the Fable lockout is not. So we show the opportunity in play, the delay exposure as a scenario, and we mark exactly where proof ends.
The honest bottom line: the economy-wide opportunity is large and measurable; the cost owed specifically to the Fable lockout is not yet quantifiable. The defensible policy concern is delay risk — which, as the previous slide shows, compounds. Sectoral figures overlap the headline and are not summed.
This isn't an abstract race. Australia starts this lockout already behind on exactly the dimension frontier AI most affects — productivity — and with a workforce still learning to use AI at all.
A country fighting a productivity problem has just been shut out of one of the most powerful productivity tools yet shipped. The cost of delay lands hardest precisely where we are weakest — and it is not abstract: it surfaces in health, cyber-defence, government services, mining and energy, small-business software and research.
A security concern can be entirely real and the response still be the wrong instrument. The honest answer is not "security was fake" — it is that a real concern is best met by controlled allied access, not indiscriminate denial.
The concern was a suspected "jailbreak." Anthropic called the vulnerabilities relatively simple and not universal; some reporting questioned the rationale.1,4,9
A worldwide, all-foreigners shutoff is a sledgehammer. It treats a trusted ally exactly like an adversary — the opposite of a targeted control.5,6
Five Eyes intelligence-sharing and AUKUS technology-sharing already rest on the highest bar of mutual assurance. The trust framework exists.8
A bounded, auditable allied channel is more secure than an all-or-nothing switch — because the alternative to access isn't safety. It's allies routing around the lockout, and capability concentrating in one jurisdiction.
In US AI-chip export architecture, Australia has been treated as a trusted, Tier-1-level allied destination, alongside the other Five Eyes nations. The framework has shifted since, but the allied-trust principle is well established.10
AUKUS Pillar II and the 2024 ITAR reforms created licence-free defence-technology sharing between the US, UK and Australia. Carving out trusted allies is a road already paved.8
The blanket order simply didn't distinguish allies. That's a fixable drafting choice — exactly the kind diplomacy resolves.5
Australia is not asking for a favour. It is asking the United States to build an AI-specific trusted-allied access mechanism — extending to model access the same allied-trust logic that already underpins AUKUS defence-technology sharing.
An access channel the US can grant with confidence because it stays in control. Three properties make it safe to say yes to:
Named models, named trusted jurisdictions (Five Eyes), defined permitted uses. Scope is explicit, not open-ended.
Named Australian entities, identity-verified users, access logging, incident reporting, defined use-cases, no onward transfer, and a single Australian-government assurance point — the governance posture AUKUS already runs on.
The US retains revocation rights under agreed conditions. A carve-out that can be withdrawn is far easier to grant than a permanent one.
Who carries it in Canberra:
On the US side, the request lands with the agencies that already run allied export carve-outs — Commerce / BIS, the State Department, the National Security Council and the White House AI office.
Australia formally requests an allied access carve-out — starting with government and security-assured industry, then broadening to the wider builder community — pursued through DFAT and the AUKUS channel, at ministerial level.
Don't wait for the request to be answered. Define the bounded-verifiable-revokable model so Australia arrives with a ready-made way for the US to say yes.
Clear guidance and support so Australian firms and researchers aren't left in limbo while access is negotiated — and a standing position for the next model, so we are never caught flat-footed again.
"Australia's relentless determination not to invest in sovereign AI infrastructure is clearly a strategic risk."
Dr Sue Keay — Director, UNSW AI Institute; founder of Robotics Australia Group; architect of the roadmaps behind Australia's National Robotics Strategy; recognised on the 2025 H20 AI 100 list specifically for sovereign-AI advocacy. She has argued Australia's National AI Plan "has all the right ingredients but lacks the cooking."11
Her existing, public, evidence-led position on sovereign AI makes her the natural voice to put this case to government. We propose her as champion — and would seek her involvement, not assume it.
Table the allied-carve-out request. The precedent, the trust framework and the economic case are all already in place. The only missing input is urgency.
Document the cost of the lockout to your work and add your voice. A quantified, named coalition is what turns a brief into a priority.
Share this. The case for access gets stronger every time someone in a position to act sees how much the delay is costing us.
Australia has now been locked out for
Prepared by GoldNet Group · June 2026 · A public, non-partisan policy brief. Facts current as at publication; all claims sourced above. The proposed role for Dr Sue Keay reflects her public positions and has not been agreed with her. Independent research and analysis — not legal or financial advice.