Sarah Breeden’s message in Bangkok on Thursday was blunt: banks can no longer count on keeping attackers out. The Bank of England’s deputy governor for financial stability told an audience at the Institute of International Finance annual meeting that lenders need to be thinking about independent recovery systems capable of restarting core operations even if their primary technology lies fully compromised.
“It’s not enough now to assume that the cyber attackers won’t get through,” Breeden said, according to a Reuters report published October 15. “We need to be thinking about what are the bare metal recovery options … that we need for the core services, and we need to think about that end-to-end across the financial system.”
Those words mark a shift in tone and priority. For years regulators pushed harder defenses, better detection, stricter patching. Now the focus widens to what happens the morning after a breach succeeds. And Breeden made clear that success looks increasingly likely when attackers wield advanced artificial intelligence.
Bare metal recovery means rebuilding systems from the hardware up, with no reliance on potentially tainted operating environments, virtual machines, or shared infrastructure. The result is an independent clean slate, ready to restore payments, settlements, or customer account access when everything else has failed. The Bank of England has begun working directly with lenders to explore these options. Results from ongoing surveys on quantum computing risks to encryption will feed into a broader sector plan.
Breeden’s October remarks build directly on ideas she laid out in a June speech at the European Central Bank Forum on Central Banking. There she questioned whether regulated firms need enhanced recovery for core systems. She asked aloud if one institution should stand ready to assume basic functions for another during prolonged outages, citing Ukraine’s “Power Banking” program launched in 2022. Or whether key players must maintain completely separate failover capabilities or the ability to rebuild compromised systems quickly from bare metal. “These are big questions, but they are the right ones,” she said in that earlier address.
A formal letter from the Bank of England to UK parliamentarians earlier this year reinforced the point. Frontier AI could drive a material increase in vulnerabilities that require rapid assessment and patching. The pace alone creates new operational dangers. Firms must plan for simultaneous multi-firm disruption and correlated failures across technology suppliers. In that context, the letter explicitly raised “bare metal rebuild” in an isolated environment or a “stand-in facility” to deliver minimum viable service. The document, sent to Dame Meg Hillier, underscores that current recovery standards may prove inadequate against faster, larger attacks.
The urgency comes from AI, which changes the economics and speed of cyber offense. Agentic systems can chain exploits, scan environments, and act with reduced human oversight. A single model might discover, test, and weaponize vulnerabilities at machine speed. Defenders lose the luxury of gradual response. The International Monetary Fund echoed these fears earlier in October, warning that an AI-enabled attack on a critical financial service could erode public confidence rapidly.
Breeden has described cyber as the risk that keeps her up at night. “What is the thing that I worry about the most is waking up and one of our most systemically important banks or a bit of core infrastructure has been attacked and is paralysed,” she told the Bangkok gathering. The basics of cyber hygiene matter more than ever, she added. Yet hygiene alone falls short when the threat model assumes penetration.
Global regulators share the concern. The European Central Bank set an October 31 deadline for significant institutions to submit updated plans addressing AI-enabled attacks, including recovery strategies. Supervisors in Britain have already pressed the largest banks to examine bare metal approaches as part of operational resilience work. In the United States and elsewhere, similar conversations gain volume. The focus moves beyond prevention toward assured continuity.
This evolution carries practical consequences for bank technology leaders. Many current disaster recovery setups depend on snapshots, replicas, or cloud instances that could inherit compromise if the attack reaches deep enough. Bare metal demands physically or logically isolated hardware, pre-configured golden images kept offline, rigorous integrity checks, and tested playbooks that assume the entire production estate requires abandonment. Cost and complexity rise, but so does assurance.
And the discussion extends further. Breeden and her colleagues have flagged risks from agentic AI in trading and payments. Autonomous agents responding identically to the same prompts could amplify market stress into full meltdowns. That worry prompted talk of market-wide kill switches or circuit breakers, concepts that once seemed extreme now enter mainstream policy debate.
Financial institutions face a dual challenge. They race to adopt AI for internal efficiency and customer service while simultaneously defending against AI-augmented adversaries. The same technology that speeds fraud detection can accelerate vulnerability discovery on the other side. Balance remains elusive.
The Bank of England plans to incorporate AI-driven scenarios into its system-wide cyber stress tests with greater emphasis on correlated failures. Coordination with industry will expand. Regulators no longer view resilience as a firm-by-firm exercise. Systemic recovery arrangements, information sharing protocols, and perhaps mutualized recovery facilities sit on the drawing board.
Breeden’s comments arrive at a moment when confidence in digital financial infrastructure faces repeated tests. CrowdStrike’s faulty update in 2024 already demonstrated how a single point of failure can paralyze operations globally.
Banks that treat bare metal recovery as a checkbox exercise will miss the point. The requirement demands genuine independence, regular rehearsal under adversarial assumptions, and integration into operational risk frameworks that reach the board level. Technology teams must collaborate more closely with risk, compliance, and business continuity functions than ever before.
Quantum readiness adds another layer. The BoE’s survey aims to gauge how prepared institutions stand for encryption-breaking advances. Post-quantum cryptography migration timelines stretch years. Recovery planning cannot ignore that horizon.
The shift is from defense-in-depth to recovery-in-assumption. Prevention still matters. Detection still matters. But the new baseline accepts that some attacks will succeed. Success then depends on how quickly and cleanly institutions can restore trusted operations without feeding tainted data back into the system.
Breeden’s intervention carries weight because it comes from the institution charged with protecting UK financial stability. Her words signal not just advice but likely future supervisory expectations. Banks should anticipate tougher questions in upcoming resilience reviews. Boards should expect deeper scrutiny of recovery time objectives when the recovery environment itself cannot be trusted.
The path forward remains under construction. The Bank of England will distill survey results and industry workshops into concrete guidance. Other jurisdictions watch closely. What emerges could reshape how the largest financial firms architect their most critical systems for the next decade.
The next step is data. “We are surveying the firms to understand where they are,” Breeden said, adding that the BoE would then “engage with industry to put a system-wide plan together.”