AI-Powered Defence, AI-Powered Threats: Navigating the Frontier AI Era

Defending against AI

United Kingdom, Jul 10, 2026

How frontier AI models are transforming cybersecurity, shrinking exploit windows and forcing organisations to rethink cyber resilience

Artificial intelligence is already reshaping how organisations operate, innovate and compete. But its next major impact may be felt most acutely in one of the business areas already under the greatest pressure: cybersecurity.

The latest Logicalis CIO Report highlights that organisations are accelerating AI investment at pace, with 94% of CIOs increasing AI spend, while many continue to grapple with governance, control and security concerns. At the same time, research from Logicalis UK&I suggests AI adoption is already creating new operational security challenges, including reduced visibility, slower incident response and emerging cybersecurity blind spots.

This tension sits at the heart of the AI security conversation. AI creates enormous opportunities to improve cyber defence, automate analysis and accelerate vulnerability management. However, it can also increase the speed and effectiveness of cyber attacks.

Recent developments around Anthropic's Claude Mythos and Project Glasswing have brought this challenge into sharp focus. Frontier AI models are now demonstrating advanced software engineering, reasoning and cybersecurity capabilities, enabling organisations to identify vulnerabilities and analyse attack paths at unprecedented scale and speed.

For business and security leaders, the message is clear: AI does not fundamentally change the principles of cybersecurity, but it does change the pace at which organisations need to operate.

What is Mythos and Why Does It Matter?

Claude Mythos is one of a new generation of frontier AI models designed with advanced reasoning, coding and problem-solving capabilities. Unlike traditional AI assistants, these models can perform complex, multi-step tasks, analyse large codebases and support security research in ways that were previously limited by human capacity.

The significance is not simply that AI can help find vulnerabilities. Security teams have used automation and scanning tools for years. The real shift is speed, scale and depth of discovery. Frontier AI models are not only identifying known weaknesses more quickly, but are also uncovering previously unknown vulnerabilities, including flaws that may have existed undetected for years or even decades within software, firmware and systems previously considered secure.

As these capabilities mature, organisations should assume that weaknesses hidden within applications, infrastructure, firmware and legacy environments may become increasingly discoverable. In simple terms, AI gives defenders greater visibility into risk, but it may also provide attackers with a much faster route to finding and exploiting weaknesses.

Project Glasswing: A Glimpse into the Future of Cyber Defence

Project Glasswing was launched by Anthropic to help secure critical software for the AI era, using Claude Mythos Preview in partnership with organisations including AWS, Apple, Cisco, CrowdStrike, Google, Microsoft, NVIDIA, Palo Alto Networks and other leading technology and cybersecurity providers.

The initiative is important because it demonstrates how frontier AI can be used defensively to identify vulnerabilities at a scale previously beyond human capability, helping organisations reduce risk before similar capabilities become widely available to threat actors. Early results were significant:

  • Anthropic and approximately 50 partners used Claude Mythos Preview to identify more than 10,000 high and critical severity vulnerabilities across critical software platforms.
  • Several partners reported vulnerability discovery rates more than 10 times higher than traditional approaches.
  • The challenge quickly shifted from finding vulnerabilities to validating, prioritising and patching them fast enough.
  • The project demonstrated that AI can dramatically accelerate defensive security activities when used responsibly.

Project Glasswing highlights a wider industry reality. AI has the potential to significantly strengthen cyber defence, but these capabilities will not remain exclusive to defenders forever.

AI Ambition is Growing Faster Than Security Readiness

The Mythos and Project Glasswing story is not happening in isolation. It is emerging at a time when organisations are rapidly adopting AI across their business. The Logicalis CIO Report found:

  • 85% of CIOs report governance as a barrier to AI adoption.
  • 79% say unchecked AI remains a serious concern.
  • 41% of UK CIOs say AI has worsened incident response times.
  • 34% report AI has already created new cybersecurity blind spots.

The challenge for organisations is no longer whether AI will impact their business. The challenge is whether their security, govenance and operational resilience can keep pace.

Technical Debt: The Vulnerability Multiplier

One of the less discussed implications of frontier AI is its ability to expose risks hidden within technical debt.

Many organisations continue to rely on ageing infrastructure, legacy applications and hardware platforms that may be approaching, or have already reached, End of Life (EOL) or End of Support (EOS). As vendor support ends, firmware updates, security patches and vulnerability fixes often cease to be available, leaving organisations increasingly exposed.

Historically, these risks may have remained hidden for years. Frontier AI changes that equation. AI-driven vulnerability research now has the potential to identify weaknesses across:

  • Legacy applications and operating systems.
  • Unsupported network and infrastructure hardware.
  • End-of-life firmware and embedded systems.
  • Previously undiscovered vulnerabilities within long-standing technology platforms.

The result is that technical debt is no longer just an operational challenge. It is becoming a cybersecurity challenge. As AI accelerates vulnerability discovery, reducing technical debt and modernising unsupported technology becomes an increasingly important component of cyber resilience.

The Opportunity: AI as a Force Multiplier for Defenders

Used responsibly, AI can significantly strengthen cyber defence. Security teams face growing attack surfaces, increasing vulnerability backlogs and continuous pressure to respond faster. Frontier AI can help organisations work with greater speed and precision by enabling:

  • Faster identification and prioritisation of vulnerabilities.
  • More comprehensive penetration testing and security validation.
  • Improved threat investigation and operational efficiency.
  • Reduced time between vulnerability discovery and remediation.

AI offers organisations an opportunity to move from reactive security towards more proactive and continuous cyber resilience.

The Challenge: Attackers Are Getting Faster Too

The same characteristics that make AI valuable to defenders also make it valuable to attackers. Frontier AI models can automate parts of the vulnerability discovery process, analyse complex attack paths and reduce the expertise required to identify exploitable weaknesses. As these capabilities become more widely available, organisations should expect exploit windows to continue shrinking. This means:

  • Less time to patch critical vulnerabilities.
  • Greater pressure on security operations teams.
  • Increased risk from exposed systems and identities.
  • Greater focus on continuous visibility and monitoring.

Cybersecurity fundamentals remain unchanged, but the operating tempo is increasing dramatically.

Why Continuous Security is Becoming Essential

As AI accelerates both defence and offence, organisations need to move beyond point-in-time security assessments.
Many businesses still rely on annual penetration tests and periodic vulnerability reviews. While valuable, these approaches may not provide the continuous visibility required in an AI-accelerated threat landscape.
Organisations should focus on:

  • Continuous testing and validation.
  • Continuous vulnerability management.
  • Continuous monitoring and response.
  • Strong identity and access controls.
  • Reduced technical debt and attack surface exposure.

The objective is no longer simply to find vulnerabilities. It is to identify them, remediate them and detect threats before they become business-impacting incidents.

How Logicalis Helps

At Logicalis, we believe the lessons from Project Glasswing are clear: organisations must become faster at both reducing exposure and responding to threats.

Reduce Exposure Before Attackers Find It
  • Penetration Testing as a Service (PTaaS)
  • Vulnerability Management
  • Cyber Risk Exposure Management (CREM)
  • Virtual Patching for vulnerabilities that cannot be immediately remediated
  • Attack Surface Assessments
  • Technical Debt and Legacy Infrastructure Risk Reviews

In situations where vulnerabilities cannot be immediately patched due to operational constraints, legacy applications or unsupported systems, approaches such as Virtual Patching and Cyber Risk Exposure Management (CREM) can help minimise exposure whilst longer-term remediation plans are developed.

Detect and Respond Faster
  • Security Operations Centre (SOC) Services
  • Managed Detection and Response (MDR)
  • Threat Monitoring and Investigation
  • Incident Response Services

As exploit windows continue to shrink, 24x7 monitoring, threat detection and rapid response become increasingly important for maintaining resilience.

Build Long-Term Cyber Resilience
  • Zero Trust Architecture
  • Multi-Factor Authentication (MFA)
  • Privileged Access Management (PAM)
  • Identity Security
  • Post-Quantum Readiness Assessments
  • Crypto-Agility Programmes
  • Certificate Lifecycle Management

Looking beyond today's threat landscape, organisations also need to prepare for future cryptographic risks. Improving certificate lifecycle management, becoming crypto-agile and beginning the journey towards post-quantum readiness can help reduce future disruption whilst strengthening security governance today.

Final Thoughts

Project Glasswing has shown what is possible when frontier AI is applied to cybersecurity. It can help defenders identify vulnerabilities faster, improve resilience and secure critical software at a scale previously difficult to achieve. However, it also provides an early indication of where the industry is heading.

AI is accelerating vulnerability discovery, increasing visibility into technical debt and changing how both defenders and attackers operate. The organisations best positioned for success will be those that continuously test, continuously monitor and continuously improve their security posture.

In the age of frontier AI, security is no longer just about protection. It is about pace.
 

Topic

Related Insights