How attacks unfold
Initial access often comes through known vulnerabilities in internet-facing systems. Frontier AI models help attackers turn newly published patches into working exploits within hours, and AI agents then scan for and exploit those weaknesses across many organisations at once, often faster than patching cycles allow. Once inside, agents harvest credentials from code repositories, file stores and secrets managers and test them across cloud and software environments in parallel.
In response, firms must rotate credentials faster than agents can use them, without disrupting services customers rely on. That forces containment decisions that may themselves cause harm, such as revoking service account credentials or taking environments offline. Those steps may also take an important business service outside its impact tolerance, with regulatory consequences.
Agents try many routes in parallel, generating far more activity to investigate before a firm can give reliable assurances about containment and data exposure. Firms are starting to use AI tools to speed that investigation, but their output can contain errors. Early accounts tend to harden into the record. Unless people who know the environment check AI output carefully, errors can find their way into risk assessments and regulatory reports and shape what the firm says long after the facts have moved on.
Why the proposals matter for Senior Managers
The proposed guidance covers governance, identification and assessment, protection, detection, response and recovery, and assurance and continuous improvement. It would apply to banks, insurers, financial market infrastructures and FCA solo-regulated firms subject to the SYSC 15A operational resilience requirements (including qualifying cryptoasset firms from October 25, 2027), but not to UK branches of overseas firms. It would not introduce new rules. Instead, it would set principles-based, outcomes-focused and technology-agnostic expectations clarifying how existing requirements on governance, operational resilience, operational risk and third-party risk apply in a cyber context, drawing on lessons from incidents and supervisory experience. Firms would be expected to refer to the NCSC Cyber Assessment Framework, which the regulators propose to use as a key supervisory benchmark. Alongside the consultation, the regulators have published two effective practices documents on managing cyber risks relating to AI and quantum computing.
The regulators say they intend to use the guidance when they supervise and enforce compliance with existing rules, so once finalised it is likely to become a reference point for supervisors assessing how well a firm and its Senior Managers assessed and mitigated cyber risk, prepared for incidents and handled them live. This includes whether Senior Managers complied with their personal regulatory obligations to take reasonable steps to ensure their business area is controlled effectively and complies with regulatory requirements relating to cyber risk, to delegate cyber risk mitigation appropriately and oversee delegated responsibilities, and to disclose appropriately information about vulnerabilities, risk mitigation and attacks of which the regulators would reasonably expect notice (the latter, and related, reporting obligations place yet more pressure on teams responding to a fast-moving incident).
This poses acute challenges for Senior Managers who must take, and oversee, containment and reporting decisions at speed before the facts are settled. The regulators say they judge reasonable steps by the circumstances at the time, not with hindsight, but any later review will inevitably start from a crystallised risk, identified weaknesses and an adverse outcome.
Many of the decisions a Senior Manager must take may have unavoidable dependencies on suppliers over which a Senior Manager has less visibility and influence in practice. Supply chain compromise remains one of the more common routes of attack, with attackers now also targeting platforms that hold access for many customers at once.
A compromise at a supplier raises two separate questions for those responding and the Senior Manager giving oversight. The first is whether to cut the supplier’s connection before the attacker can use it to reach the firm, a decision that usually must be taken on the supplier’s information while it is giving every customer the same holding statement. The second is how long the firm’s own services can run without that supplier, which depends on how quickly the supplier can recover. Supervisors have criticised reliance on untested assumptions about suppliers’ cyber recovery, and in one case the PRA found that a Senior Manager who relied on supplier confirmations that were partly statements of intent, as opposed to genuine assurances, had not taken reasonable steps to ensure compliance with applicable regulatory requirements. The draft guidance confirms that a firm and its senior management remain accountable where they rely on third parties, and expects firms to test whether recovery arrangements remain effective where dependencies are compromised, degraded or unavailable. Contingency planning for the unavailability of important suppliers is now a critical part of a firm’s resilience.
Alongside active and engaged oversight, maintaining robust audit trails of what was known, challenged and decided, and why, is key. Meeting minutes, notes and emails are all helpful, particularly if they demonstrate critical interrogation of management information and a timely response to external developments, such as new threat intelligence.
Oversight and controls expectations
In previous enforcement action the FCA has already indicated that boards are ultimately responsible for ensuring cyber controls meet standards of resilience, should set an appropriate cyber risk appetite and should ensure that response plans are well rehearsed. Under the regulators’ operational resilience rules, boards must approve and regularly review their firm’s important business services, impact tolerances and self-assessment. Board reporting should show whether controls work against current threats, including AI-enabled ones, and minutes should adequately record challenge and debate among board members, as well as what decisions were made as a result. Management information is key—with an emphasis on equipping boards and Senior Managers to critically assess and challenge risk assessments, remediation plans and the extent to which fixes mitigate exposure in practice. The proposed guidance makes the board accountable for the firm’s cyber governance and expects board management information to cover, among other things, backlogs of unpatched vulnerabilities, privileged access and third-party cyber risks, while boards and senior management should set a risk appetite on key control indicators for vulnerability management.
Those expectations require clear senior oversight and escalation paths. The threat landscape now calls for nuanced senior-level decision-making on a rolling basis, which means that effective processes for the identification and timely escalation of key information are essential. Senior decision-makers should pay particular attention to challenging whether controls adequately address existing and emerging cyber risks. They should also focus on their capacity to find and fix vulnerabilities and tighten identity and access controls quickly without causing new problems. In practice, decisions about the speed of remedial work can be especially fraught. Move too slowly and the risk of successful attack increases. Move too quickly and the cure can cause its own harm: with insufficient time to assess and mitigate impacts on dependencies, patches may themselves disrupt essential services.
Escalation at inference speed
It has long been challenging to implement—and test—crisis management procedures that are sufficiently rapid, robust and flexible. Inference-speed attacks leave even less room for delay. Firms should settle in advance who can authorise containment steps that may breach impact tolerances, how fast decisions can be made out of hours, and when the board is told. The draft guidance expects clear decision-making authority during an incident, including where information is incomplete or rapidly changing, asks firms to assess, document and oversee the risks of containment steps such as taking core systems offline, and encourages them to consider automation to contain incidents before they propagate. Firms should also be clear about who decides whether an incident meets regulatory reporting thresholds, including under the new operational incident reporting requirements from next March.
A Senior Manager’s duty under Senior Manager Conduct Rule 4 to disclose appropriately any information of which the regulators would reasonably expect notice makes regulatory reporting a personal obligation too. The draft guidance proposes that the SMF24 (Chief Operations) should ordinarily have overall responsibility for implementing the firm’s cyber risk and resilience strategy and reporting on it to the board, although firms may allocate this to another suitable SMF, such as the Chief Risk Officer (SMF4). It also treats as a major cyber incident any incident reportable under the new operational incident reporting rules taking effect on March 18, 2027, or of which the regulators would expect notice under Principle 11 or PRA Fundamental Rule 7.
The wider context
Alongside these proposals, boards of financial services firms face two related government initiatives. The voluntary Cyber Governance Code of Practice, to which the draft guidance cross-refers, asks boards to gain assurance that cyber controls work and, in an incident, asks directors to take responsibility for their individual regulatory obligations, such as reporting, and to support the organisation in critical decisions and external communications. The Cyber Resilience Pledge goes further, since signatories commit publicly to implementing every action in the Code. For firms that have not signed, the Code is still likely to inform how board oversight is judged after an incident. For those that have, the public commitment will sit alongside what supervisors expect of Senior Managers.
Preparation is key
Rehearsing against realistic AI-enabled scenarios will help firms respond and prepare. Those exercises should test governance as well as technical response: whether board reporting supports real challenge, whether escalation and decision rights work at speed, and whether responsibilities maps reflect who would actually take the critical decisions. The draft guidance also expects response and recovery testing, where proportionate, to go beyond traditional tabletop exercises.
The consultation closes on February 26, 2027, after which the regulators expect to publish a policy/feedback statement including the final guidance.
The regulators also invite comments on how their AI and quantum effective practices documents could be improved, which is where firms’ practical experience of AI-enabled risk mitigation – and responding to AI agent attacks – will really count.
For more, read our cybersecurity series on building cyber resilience in an evolving threat landscape.