A body of survey data on AI cybersecurity in financial services is converging on a consistent problem: institutions know the threat is growing, but many are not ready for it. The latest addition to that picture comes from Cisco, whose report Relentless Defense: A global report about cybersecurity in the AI era draws on responses from 8,000 security leaders across 30 markets.
The report does not publish its full financial services breakdown publicly in the version referenced here, but Cisco is presenting its findings at Open Banking Expo UK & Europe, where Rob Lay, the company’s UK&I chief technology officer, is scheduled to speak on ‘Building trust and security into the AI powered Infrastructure’ at 10.35am on the Main Stage on Day Two.
What other surveys say about AI cybersecurity in financial services
Cisco is not alone in documenting the problem. Separate research from PwC found that 51% of security leaders in financial services rank attacks targeting AI systems among the top cyber threats their organisations are least prepared to address. That figure deserves a moment’s pause. It is not that respondents are unaware of AI-related threats; it is that they consider themselves under-equipped to handle them.
The gap between awareness and readiness is a thread running through several data sets. PwC’s framing is notable for what it implies about resourcing and prioritisation: organisations are acknowledging a class of threat they have not yet built adequate defences for.
Prompt injection, deepfakes and the operational reality
The Thales financial services data threat report adds more granular detail. According to Thales, 70% of FinServ organisations expressed concerns about rapid change in AI ecosystems, which respondents identified as their top AI security risk. The pace of change itself, rather than any single attack vector, is what is generating the most anxiety.
Thales also reported that 64% of financial services organisations surveyed had suffered prompt injection attacks on their AI applications, and 62% had experienced sensitive data disclosure. These are not theoretical exposures. Prompt injection, which involves manipulating the inputs to an AI model to make it behave in unintended ways, has moved from a research curiosity to an operational concern for a majority of firms in the sector, according to the Thales data.
Deepfakes represent a separate but related threat. Thales found that 60% of financial services organisations had experienced deepfake attacks, with 50% reporting that AI-fuelled attacks had resulted in reputational damage. The combination of synthetic media and financial services, a sector where identity verification and executive communications carry genuine weight, creates a reasonably clear attack surface.
Taken together, the Thales numbers suggest that AI cybersecurity in financial services is not primarily a forward-looking concern about what might happen. For a large proportion of the firms surveyed, breaches and attacks have already occurred.
What the Cisco session is claiming to address
Cisco’s session at Open Banking Expo is framed around ‘building trust and security into AI powered infrastructure’, which is broad enough to cover both the architectural questions and the governance questions that the survey data raises. The claim that infrastructure trust can be built and maintained while AI systems evolve rapidly is one that the numbers from PwC and Thales complicate rather than resolve.
The honest version of the problem, as the data describes it, is this: financial institutions are deploying AI faster than they are developing defences specific to AI-native threats. Prompt injection attacks and deepfakes are not extensions of older attack types; they require different detection approaches, different controls, and different incident response frameworks. The 70% of FinServ organisations that Thales found concerned about rapid AI ecosystem change are, in effect, describing a situation where the attack surface is expanding faster than the security perimeter.
Whether Cisco’s session will move beyond the well-documented diagnosis to something more concrete is a question the Open Banking Expo audience will be in a position to judge on Day Two. Lay’s session is scheduled for 10.35am on the Main Stage.
