Key Takeaways
- Lloyds’ chief security officer Matt Rowe believes agentic AI can give banks a “home‑field” edge by spotting security gaps before attackers exploit them.
- A recent cyber‑attack simulation hosted by Lloyds, Google and Hack The Box let over 30 UK financial‑services teams practice real‑world defence and sharpen attacker‑think mindset.
- Rowe warns that the cyber‑defence burden is growing, making advanced tools essential for staying ahead of increasingly sophisticated threats.
- Anthropic’s Claude Mythos Preview, though currently limited to a few US banks and tech firms, can uncover deep‑rooted vulnerabilities across major OSes and browsers, signalling a new wave of AI‑driven risk discovery.
- Some security professionals view Mythos’s restricted rollout as marketing hype, but many agree banks must accelerate AI adoption to outpace criminal hackers.
- AI automation of routine ticket‑patching and alarm‑investigation frees security staff to focus on harder problems, such as detecting stealthy adversary behaviour that mimics normal activity.
- Rowe sees Mythos as just the beginning; future AI models will likely improve vulnerability detection further, making this an exciting era for cyber‑security professionals.
Agentic AI as a Defensive Advantage for Banks
Lloyds’ chief security officer Matt Rowe argues that banks possess an inherent “home‑field” advantage because they understand their own technology stacks and processes better than any external attacker. However, this insight has not always translated into stronger cyber defences. By deploying agentic AI—intelligent systems that can autonomously scan code, configurations, and logs—banks can turn that internal knowledge into proactive threat hunting. Rowe explains that large language models excel at finding a “needle in a haystack,” allowing security teams to uncover hidden attack paths that would otherwise go unnoticed. If these vulnerabilities are patched before hackers gain a foothold, banks can effectively stay a step ahead in the relentless cyber‑arms race.
Exercise Demonstrates Real‑World Cyber Defense Practice
To illustrate the potential of such proactive defence, Lloyds teamed up with Google and the cyber‑readiness platform Hack The Box to host a two‑day simulation exercise earlier this week in London. More than 30 teams drawn from across the UK financial services industry participated, each tasked with defending realistic network environments against sophisticated, live‑fire attacks. The exercise was designed not only to test technical skills but also to encourage participants to “flex their attacker muscles”—thinking like adversaries to anticipate moves and refine defensive strategies. Rowe noted that the burden of maintaining strong defences has grown significantly, and these high‑fidelity drills provide a rare opportunity to practice under pressure while learning from peers across the sector.
The Growing Burden and Need for Advanced Skills
Rowe emphasized that the cyber‑defence workload for banks has become increasingly heavy, with security teams constantly juggling ticket management, patch deployment, and alarm investigation. He observed that the pace of threat evolution does not make the job any easier, which is why exercises like the Lloyds‑Google‑Hack The Box simulation are invaluable. By immersing defenders in realistic attack scenarios, banks can develop deeper intuition about attacker behaviour and improve their ability to respond swiftly. The ultimate goal, Rowe said, is to move beyond reactive firefighting toward a stance where vulnerabilities are identified and remediated before they can be weaponised.
Emerging AI Models Like Claude Mythos Preview Pose New Threats and Opportunities
A focal point of the discussion was Anthropic’s Claude Mythos Preview, a new large language model capable of identifying software vulnerabilities across every major operating system and web browser—including flaws that have remained undetected for years. Although Anthropic has initially restricted Mythos’s rollout to a handful of US banks and technology firms, labelling it “too dangerous” for broader release, Rowe described the model as a “step forward.” He believes Mythos will enable banks to uncover weaknesses they would previously have missed, thereby tightening their defences. However, he also cautioned that any vulnerabilities uncovered must be patched at “high velocity,” given the speed at which attackers can weaponise newly discovered flaws. Rowe views Mythos as merely the first of many forthcoming AI models that will likely perform similar—or even better—vulnerability discovery tasks.
Skepticism and Calls for Greater AI Adoption
Not all participants shared Rowe’s enthusiasm. Some cyber‑security specialists attending the simulation characterised Mythos’s limited availability as “marketing hype,” questioning whether the model truly delivers breakthrough capabilities beyond existing tools. Nevertheless, a common thread emerged: banks must accelerate their adoption of AI to stay ahead of criminal hackers. One ethical hacker involved in the exercise asserted, “With AI, we can improve our defences. If we can keep up the pace and not stay behind with the technology, we can actually win.” This sentiment reflects a growing consensus that, while caution is warranted regarding AI’s potential misuse, the technology’s defensive benefits are too significant to ignore.
Automation of Routine Tasks Enables Focus on Sophisticated Threats
Rowe highlighted another practical advantage of AI: the automation of mundane, repetitive security tasks. Traditionally, analysts spend considerable time opening and closing tickets, applying patches, and investigating low‑level alerts—activities that, while necessary, consume valuable human effort that could be directed toward more complex challenges. By delegating these routine functions to AI agents, security teams can concentrate on “harder problems” such as detecting adversaries who blend in with normal user or system behaviour to evade detection. Rowe described this shift as making the job “more interesting” and noted that professionals in security “just want to do a brilliant job.” AI now provides the tools to accomplish feats that would have seemed unimaginable only a few years ago.
Future Outlook: Continuous Evolution of AI in Cybersecurity
Looking ahead, Rowe expressed optimism that the current wave of AI‑driven security tools is only the beginning. He anticipates that models comparable to or superior to Claude Mythos Preview will emerge, continually enhancing the ability to find and remediate vulnerabilities at scale. As AI capabilities advance, banks will need to adapt their processes, ensuring that patch management and threat response keep pace with the velocity of discovery. Ultimately, Rowe views this period as an “amazing time” to be in cyber security, where the fusion of deep institutional knowledge and cutting‑edge AI can transform defensive posture from reactive to genuinely proactive. The ongoing collaboration between financial institutions, technology providers, and platforms like Hack The Box will be crucial in shaping a resilient cyber ecosystem for the years to come.