Five Eyes to boards: AI has shrunk the cyber clock to months, and Australia's ACSC co ...

Five cyber agencies, one message: get the basics right, wire AI into your defence, and stop treating a breach as a maybe. Here's what the warning means for Australian organisations, and what the security industry makes of it.

Overnight, the cyber security agencies of the Five Eyes alliance, Australia, the United States, the United Kingdom, Canada and New Zealand, put out a rare joint call to action . The thrust is blunt. AI is rewriting cyber risk faster than most boards are moving, and the time to act is now.

The line that should stop executives mid-coffee: "The timeline is not years, it is months."

Australia's signature is on it. The Australian Signals Directorate and its Australian Cyber Security Centre co-signed alongside the NSA, CISA, the UK's NCSC, Canada's CCCS and New Zealand's NCSC.

A rare joint warning, and a tight deadline

Joint Five Eyes advisories aren't new. Six agency heads putting their names to a single page one technology is.

The signatories are ACSC head Stephanie Crowe (ASD), Rajiv Gupta (Canada's CCCS), Catriona Robinson (New Zealand's NCSC), Richard Horne (the UK's NCSC), David Imbordino (NSA) and Nick Andersen (CISA). That's the entire alliance speaking with one voice on the same day.

Their core claim is speed. AI "lowers barriers for malicious actors and increases the speed and complexity of attacks, shrinking the window between vulnerability discovery and exploitation ever more quickly," the statement reads. The same models that help attackers move faster also, they note, "offer powerful tools to strengthen defence."

The agencies are careful to frame this as a leadership problem that sits with boards and executives. In their words, "cyber risk can no longer be treated as a purely technical issue. This is a core business risk and leadership responsibility." Having controls isn't the bar. Leaders need to be confident those controls "will perform during a real incident."

What the agencies actually want from leaders

The statement asks executives to do four things: understand and assess risk, readiness and accountability; prioritise foundational security practices and controls; give cyber leaders real authority and resources; and stay engaged as the threats and the guidance keep shifting.

There's a warning baked into the close. Success "will not come from having the most tools," the agencies say. "It will come from getting the basics right, acting quickly, and integrating cyber security into core business strategy." The organisations that don't, they add, "will face growing operational and strategic disadvantage."

Five practical moves they say are now urgent

The agencies list five actions they describe as not new, but now urgent.

Reduce your attack surface by limiting unnecessary system access and external connectivity, and isolating anything that doesn't need to be exposed. Accelerate patching, because AI is compressing the gap between a flaw going public and someone using it. Deal with legacy systems, which the statement calls "not just technical debt, they are strategic liabilities." Tighten identity and access controls with strong authentication and regular permission reviews. And prepare for incidents before they land, on the assumption that breaches will happen.

Then the part that gives the document its edge: use AI to defend. "Adversaries are already using AI to move faster and more effectively. Defenders must do the same."

ACSC head Stephanie Crowe struck an upbeat note in the Australian release. "I'm actually really positive that we have the tools and we have the capabilities," she said, adding that organisations that take the time to look hard at their cyber risk plans are "in a really good place," especially if they learn to use AI defensively before adversaries get further ahead.

Why this hits differently in Australia

A warning from overseas is easy to file under "later." A warning your own signals directorate signed is harder to ignore.

This lands while the ASD is already tightening the screws at . The directorate recently updated its Information Security Manual , the rulebook government agencies must follow, to make secure-by-design and secure-by-default standard practice rather than nice-to-have. The ACSC's existing AI security guidance is where the agency is pointing Australian firms .

For local organisations running stretched security teams, ageing infrastructure and scattered environments, the speed problem is the real sting. If your detection and response is still mostly manual, minutes-to-exploit is a genuinely uncomfortable number.

The Fable 5 and Mythos backdrop

The timing isn't an accident. The warning follows months of argument over frontier models that can do offensive cyber work.

In recent weeks the US imposed export controls on Anthropic's most capable model, Fable 5, requiring a licence for non-Americans to use it. Anthropic then withdrew Fable 5 entirely. It shares its underlying model with the restricted Mythos, the version that set off the cyber-capability alarms in the first place. Security leaders pushed back in an open letter , arguing the curbs hurt defenders while adversaries with near-equivalent models keep closing the gap. DarkReading covered the backlash here , and SecurityWeek rounded up industry reaction here . The Guardian's piece on the national-security framing is here .

That fight is the subtext of "months, not years." When the gap between a locked-down frontier model and its freely available equivalent is measured in a couple of quarters, the agencies' deadline starts to look conservative.

What the security industry makes of it

The statement landed in inboxes with a wave of vendor reaction. Sorted by who said what, here's the signal.

Elastic ANZ Country Manager Jeremy Pell read it as a direct message to Australian organisations. "Australia's own ACSC has co-signed this call to action, and that carries weight. This isn't a theoretical risk assessment from overseas. It's our own cyber security leadership telling boards and executives that the threat has fundamentally changed in nature and speed," he said. Pell argued the fix is an agentic security operations model, "where AI handles triage and enrichment at machine speed while human analysts focus on judgement and escalation," and noted Elastic works with 45% of the ASX top 20 and 50% of the Fortune 500 on security, and observability. Elastic competes in that security-analytics and SIEM space with Splunk (now Cisco ), Microsoft Sentinel and CrowdStrike .

Arctic Wolf APAC Director David Hayes said the warning reinforces what security leaders already see. "AI is making it easier for cybercriminals to launch attacks faster, at greater scale and with less effort than ever before," he said. His sharper point is that attackers often don't need new techniques, they just need to find weaknesses faster than defenders fix them. He pointed to Arctic Wolf's State of the Cybersecurity Attack Surface Report, which found one in three IT assets is either missing a critical control or misconfigured. The managed detection and response specialist, which employs around 70 staff locally and counts Arts Centre Melbourne, the Parramatta Eels and Brighton Grammar School among its Australian customers, plays against CrowdStrike, SentinelOne and Expel in the MDR market.

TrendAI Federal Government Lead for ANZ Martyn Beal read it as a board-level wake-up call with a practical core. TrendAI is the enterprise security unit Trend Micro rebranded in March , and it lines up against CrowdStrike , Microsoft and Palo Alto Networks in enterprise AI security.

"The Five Eyes cyber agencies have issued a rare and timely warning: AI is collapsing the time between vulnerability and exploitation from years to months. What once took years can now happen in a matter of months, fundamentally changing the risk landscape for government agencies, critical infrastructure operators and businesses alike. Cybersecurity can no longer be viewed as solely an IT issue. It is now a Board-level and Executive Leadership priority because the consequences of a successful attack extend far beyond technology, impacting operations, reputation, public trust and national resilience," Beal said.

"The agencies' advice is refreshingly practical, get the basics right: shrink your attack surface, patch faster, retire legacy systems, tighten identity controls, and assume breaches will happen. But they add a sharper edge: defenders must use AI as deliberately as attackers do. There are four core principles that ground a robust AI security strategy: gaining visibility into AI usage, systems and agents interact across environments, understanding the context and intent behind those interactions, enforcing policy and control over usage and agent-driven actions, and introducing human oversight at critical decision points.

"The organisations that will succeed are not necessarily those with the most security tools, but those that can move fastest to identify risk, respond to threats and build resilience. In the age of AI, speed has become one of the most important security controls."

Illumio Public Sector CTO Gary Barlet , a retired US Air Force lieutenant colonel and former US Postal Service OIG CIO, took aim at the idea that restricting one model buys safety. "The idea that the threat was somehow going to be slowed because models like Mythos weren't broadly released was always wishful thinking. Whether it's Mythos, Fable, or the frontier model, it isn't a matter of if these capabilities become widely available; it's when," he said. His blunter line is for the patch-and-pray crowd: "It's time for organisations to stop treating a breach as a possibility and start treating it as an inevitability." Illumio sells microsegmentation, or Zero Trust Segmentation, against rivals like Akamai's Guardicore and Zscaler .

Invi Grid CEO Yogita Parulekar reached for the heaviest historical parallel going. "We are at an inflection point: advanced autonomous AI models can be used for mass surveillance, autonomous weapons or attacks at a scale and speed never seen before. This moment in history parallels the debate over building the atomic bomb, with widespread consequences that will be hard to contain," she said. Her practical advice: review your infrastructure for public exposure points, your applications for attackable vulnerabilities, and your AI for "excessive agency that malicious actors can turn into renegade hackers." She closed with the article's tightest summary of the deadline: "The window to get ahead is now and it is short."

Invi Grid Head of Customer Success and Partnerships Kristen Santora made the case for looking inward rather than only at attackers. "The instinct is to look outward at attackers using AI as a weapon. The harder conversation is looking inward at the AI infrastructure already running inside your organization and whether it was built with the governance and security controls to withstand what is coming," she said. Her warning on model risk is specific: "A model like Fable 5 can be jailbroken to exploit cybersecurity vulnerabilities, and when that happens, every organization that integrated it inherits the exposure." Safeguards added after the fact, she argued, are not enough.

Finite State Head of Policy and Compliance Doc McConnell , who previously shaped federal cyber policy at CISA, put the onus on makers and users of connected kit. "If you're making software or connected devices, then you should assume that AI tools can find vulnerabilities in your products. Don't outsource vulnerability discovery to bad actors. Find and fix the problems internally first," he said. He also wants the access equation flipped for defenders: governments and AI labs, he argued, "must accelerate access to these advanced AI models by defenders, abandoning 'pilot access' and 'trusted partner' programs that give attackers the advantage." Finite State works in product and firmware security and software supply chain, the SBOM-heavy corner of the market.

Black Hills Information Security owner John Strand brought insider weight. "After spending 15 years in the Five Eyes intelligence community, I can say with confidence that we are horribly behind in understanding how quickly AI can move through the entire attack lifecycle," he said. The penetration-testing veteran warned that the community still treats many of these capabilities as theoretical "while AI is already accelerating everything from reconnaissance and target discovery to exploitation and post-compromise operations," and that long-held assumptions warning time and attacker effort "are rapidly becoming outdated."

Horizon3.ai Vice President of Sales for APJ Gareth Cox made the case that speed is only half the story. The autonomous-pentesting firm, whose NodeZero platform competes with Pentera and Cymulate , argues the win is proving which flaws can actually be used. "Five Eyes is right: AI is compressing the time from flaw discovery to exploitation. However, the real problem is not finding more bugs; it is validating which weaknesses are actually exploitable in an IT environment. Enterprises needs to deploy solutions which can support them to continuously prove what is exploitable in their real environment, show how attacks chain across web, identity, cloud, and infrastructure, and focus teams on the fixes that measurably reduce business risk. This includes continuous, production-safe autonomous pentesting across infrastructure, cloud, identity, and external attack surfaces, rather than periodic assessments," Cox said.

"Ultimately, Australian CISOs and security leaders require attacker-validated evidence they can stand behind with boards and auditors," Cox concluded.

Xcape, Inc. COO Noelle Murata named the open-source clock directly. "The core threat here is that Anthropic's Fable 5 and OpenAI's Daybreak models lower the barrier for bad actors while exponentially increasing the speed, scale, and sophistication of attacks. With a 6- to 8-month lag between frontier models and their open-source equivalents, destructive capabilities will soon be public," she said. The pentest and incident-response firm's COO laid out four takeaways for leaders:

Fix the core vulnerabilities AI will exploit: legacy systems, slow patching, unnecessary internet exposure, weak identity controls, no incident planning.

Treat cyber risk as a whole-of-organisation business issue, not an IT afterthought.

Plan for rapid obsolescence as restricted AI becomes free within months.

Double down on the basics, because success comes from fundamentals, speed and integrating security into core strategy.

Her closer doubles as a memo to every board: "Tomorrow's nation-state capability is quarter's open-source download. Fix the basics now or explain later why you didn't."

The dissenting view: fear, or overdue buy-in?

Not everyone reads the statement as a fire alarm. Suzu Labs Managing Director Steven Swift , who runs an offensive-security and adversary-emulation shop, called parts of it overcooked.

"This is just fearmongering around offensive AI capabilities," he said, arguing that APT groups and pentesters already run a "huge and well-documented" volume of offensive campaigns with existing tooling. AI, in his telling, is a faster way to run the same kill chain security teams have always faced. "There is no unique, magic hack that AI has access to. AI agents are simply a different way of running the same exact kill-chain as we would have without them," he said.

His real target is underinvestment. Organisations get compromised, he argued, because they "famously underinvest in cybersecurity," and the fix is unglamorous: secure configurations, timely patching of everything, and properly licensed security tooling. There's a twist in his conclusion, though. "If mentioning AI security suddenly gets more buy-in, then that's a win for finally making meaningful improvements to security posture."

Which is the useful tension in this whole conversation. Swift and the agencies actually agree on the prescription, get the fundamentals right, even if they disagree on how scared you should be. Strand, from inside the intelligence world, thinks the fear is warranted. Murata thinks the maths of open-source diffusion settles the argument. Read together, they're split on how alarmed to be while landing on the same to-do list.

Strip away the model names and the deadline, and the agencies are asking for things security teams have begged boards to fund for a decade: patch faster, kill legacy, lock down identity, assume you'll be breached, and rehearse the response.

What's changed is the cost of dragging your feet. When the gap between a vulnerability and a working exploit is measured in minutes, and the gap between a locked-down frontier model and a free one is measured in months, "we'll get to it quarter" becomes a strategy with a body count.

Australia's cyber leadership signed its name to that timeline. The homework is the boring stuff, done quickly, with AI wired into the defence rather than tacked on at the end. The organisations that treat this as a board problem this quarter will be fine. The ones waiting for a neater moment are the case study someone writes year.

The clock the agencies described is already running. It started before the statement hit your inbox.