Cyber protection against advances in frontier AI models

A reality check

Looking at what Claude Mythos achieved, Rik Ferguson, vice-president of security intelligence at Forescout, notes that the AISI analysis shows that before Mythos, no AI model had completed a 32-step simulated corporate attack chain end-to-end.

“Mythos Preview did so in three out of 10 runs. GPT 5.5 did so in two,” he says. “The capability gap between the two leading frontier models is narrower than the coverage implies. The governance gap is considerably wider.”

IT security experts acknowledge that organisations with a robust cyber security strategy are in a far better position than those with less robust cyber defence capabilities.

“The vulnerability pipeline was never the core problem for defenders. We were never suffering from a shortage of things to worry about,” says Ferguson. “We already have more disclosures, more advisories, more proofs of concept, and more exposure data than most organisations can realistically operationalise,” he adds.

However, what Mythos does, according to cyber security experts, is compress the timeline between weakness, discovery, weaponisation, and the need for defensive action.

“A single adversary can now automate reconnaissance, generate exploit variants, analyse source code, weaponise misconfigurations, and adapt phishing or social engineering campaigns at machine speed,” says Aditya K Sood, vice-president of security engineering and AI strategy at Aryaka.

Sood says the challenge for CISOs is that they are no longer tackling just an “AI adoption risk” but the rise of AI-amplified adversaries capable of iterating faster than traditional defence cycles.

A board-level problem

This has implications for how organisations react to cyber security incidents. PA Consulting’s Atkinson urges executive teams to focus on how quickly controls operate in practice, particularly patching for internet-facing and identity and access management systems.

Most organisations have built vulnerability management processes around a manageable number of disclosures, addressed through governance, risk assessments and change windows. But, as Atkinson points out, in the near term, AI will dramatically increase the rate at which vulnerabilities are found. “Organisations may face hundreds of new vulnerabilities across legacy systems within weeks, far exceeding existing change capacity,” he says.

Frontier AI is not removing the importance of cyber fundamentals – it is raising the cost of failing to deliver them at speed

“Security failures are increasingly likely to result not from lack of awareness, but from inability to act quickly on what is already known. Frontier AI is not removing the importance of cyber fundamentals – it is raising the cost of failing to deliver them at speed,” he adds.

Among the challenges organisations can face is the visibility gap, which Atkinson says is where some organisations lack a complete understanding of assets, dependencies and risk exposure. This makes prioritisation difficult.

Another problem is what he calls “governance bottlenecks”, whereby “most vulnerability assessment and remediation processes are not designed for high-volume remediation”.

Atkinson recommends that CISOs consider adapting their cyber security strategy to support patch deployment at speed and scale, especially for internet-facing systems. This, he says, also needs accurate asset and dependency management, to support prioritisation.

Atkinson also urges organisations to streamline change governance, so they can operate under continuous update conditions. At a senior management level, he says CISOs must ensure there is clear ownership of risk decisions when trade-offs must be made quickly. “Without these, even the best detection tools or AI-assisted defences will struggle to compensate,” he warns.

Surviving an AI-powered cyber attack

Forescout’s Ferguson recommends that organisations shift their focus towards operational survivability: preserving visibility, constraining attacker manoeuvre space, limiting blast radius, and maintaining continuity under stress.

“The organisations that can patch at pace without the wheels falling off will be the ones that have already done the foundational work on asset inventory, segmentation and prioritisation based on actual exposure,” he says.

In the frontier AI era, Ferguson believes operational survivability is the measure that matters. “The organisations that understand that now won’t be the ones scrambling up the beaches when the patch wave starts to build,” he says.

From his own work, Ferguson says current AI models are capable of exploiting what he calls pattern bugs. These are the injection flaws, leaked secrets, known bad dependencies and chaining findings across systems. However, he says: “Where they still fall short is anywhere correctness depends on intent.”

He notes that business logic and authorisation flaws remain the category where AI models are consistently weakest. “Unlike pattern bugs, they require understanding what code is supposed to do, not just what it does. That gap hasn’t yet been closed,” he says.

In fact, Ferguson believes AI coding tools could widen the gap further. “Vibe coding guarantees we don’t hit a plateau: higher tempo development, more dependencies, more confident shipping,” he says.

The volume of code being written with AI assistance means the target is expanding at the same time as the tools for finding coding flaws are improving

But while the number of defects on each piece of code produced may remain more or less consistent, Ferguson warns that the total attack surface area grows. “The volume of code being written with AI assistance means the target is expanding at the same time as the tools for finding coding flaws are improving,” he says.

Given the attack surface is rapidly expanding and the ability to direct attacks at vulnerable systems is being accelerated thanks to advances in frontier AI models, Aryaka’s Sood recommends that CISOs focus on the patch and response process, which he says needs to assume that a critical vulnerability may be weaponised within 24 hours of disclosure, or even sooner.

“Relying on slow patch cycles, manual triage, or periodic security reviews is not viable when adversaries can automate reconnaissance, weaponisation and exploitation at machine speed,” he says.

Sood advises CISOs to develop rapid-response security models that include pre-positioned response playbooks, AI-assisted prioritisation and resilient architectures capable of quickly containing compromise. “In practice, CISOs must assume that once a weakness becomes visible, AI-enabled adversaries can rapidly operationalise it before traditional defences can react,” he adds.

His advice to IT leaders is to “stop planning for the attacker you knew and start planning for the attacker that frontier models enable”. 

As Sood points out, an AI-powered attack is faster, more contextually aware, more persistent and more scalable than anything the security industry has faced. “CISOs who adapt most quickly to manage the AI attack surface will lead enterprise security in the frontier model era. Those who treat this as an incremental update to existing frameworks will find that the gap between their defences and the threat has quietly become insurmountable,” he warns.

Rethinking security tooling

Looking at the cyber security tools market, Ferguson says vulnerability management and quality assurance are converging as AI tooling improves.

However, he adds: “The question each answers remains distinct, and no amount of converging changes that. Quality assurance asks whether something works. Vulnerability management asks whether it can be abused and what the blast radius looks like.”

When models like Mythos become more widely available, Ferguson predicts there will be a spike in disclosed vulnerabilities, and then cyber security tools providers will confront the problem.

“The question is whether discovery translates into remediation or just accumulates as a bigger backlog,” he says. “The pressure to ship doesn’t disappear because a model found more bugs. Without hard blocks for exploitable, high-impact issues and firm deadlines for everything else, the surge risks becoming the new normal rather than a more durable correction.”

Ferguson believes human judgement is irreplaceable in the research and security pipeline. “At Vedere Labs, we already use Claude Opus 4.6 in our research workflow and have reported several zero-days found through that process,” he says. “The goal is turning faster research into better protection.”

It is clear that the tech sector is now in an arms race as advanced AI models continue to evolve. For IT chiefs, Claude Mythos Preview is a wake-up call, and cyber security teams will need to shore up corporate IT defences against the imminent threat posed by AI-accelerated cyber attacks.