The financial sector has moved past the phase where the main question was whether to adopt artificial intelligence. The debate has shifted. Now, the focus is on how to govern systems that are already entering production, gaining autonomy, and beginning to operate over sensitive data, processes, and decisions. A new report from the Cloud Security Alliance, commissioned by Anjuna, shows that banking, insurance, and other financial services are rapidly moving towards agentic AI, but they still carry a concerning lack of visibility over their risks.
The study, State of Cloud and AI for Financial Services 2026, based on 340 responses from professionals in cloud, artificial intelligence, cybersecurity, compliance, and risk management across financial organizations worldwide, depicts a sector that no longer treats AI as a laboratory experiment. According to the report, 62% of surveyed organizations have already deployed AI agents. At the same time, 20% acknowledge experiencing security incidents related to AI, and another 21% are unsure if they have had any.
Agentic AI is already within financial institutions
The clearest data point from the report is that adoption is advancing. Only 27% of organizations reported not using AI agents. Over a third, 35%, indicated they are already implementing them in production, while another 9% are in an advanced adoption phase. Additionally, nearly half of the respondents, 49%, say they are exploring or launching pilot programs.
The difference from previous years is significant. AI is no longer limited to internal analysis, simple chatbots, or low-risk automation. Agents are beginning to appear in customer service, cybersecurity operations, back office, and fraud detection. These are areas where efficiency improvements can have a direct impact, but also where failures can affect clients, operations, sensitive data, or regulatory compliance.
| Report Indicator | Key Data |
|---|---|
| Organizations that have already deployed AI agents | 62% |
| Organizations not using AI agents | 27% |
| Active implementation in production | 35% |
| Advanced adoption | 9% |
| Exploring or pilots | 49% |
| Known security incidents related to AI | 20% |
| Organizations unsure if incidents occurred | 21% |
| Use of any cloud services | 98.3% |
| Mainly or entirely cloud architectures | 33% |
| Moderate or strong senior management support | 91% |
Use cases indicate where transformation begins. Customer service tops the list at 63%. Followed by cybersecurity operations with 47%, back office with 44%, and fraud detection with 41%. These processes can benefit from automation that saves time, enables quicker responses, and manages large volumes of information, but they also require control and traceability.
The report highlights a particularly delicate point: 93% of organizations using agents have granted them some level of autonomy. This does not mean all agents can make critical decisions independently, but it indicates that AI is entering a stage where it doesn’t just recommend but also begins to act within workflows.
Autonomous payments: a near frontier
One of the most striking conclusions is the expectation regarding payments managed by agents. 85% of respondents believe AI agents will initiate and execute payments on behalf of consumers. This is not a minor hypothesis. In finance, allowing an autonomous system to move money requires redefining authorization, identity, consent, limits, auditing, and responsibility.
Most respondents seem aware of this challenge. 65% believe autonomous payments will require a new authorization model. This could include more granular permissions, amount limits, contextual validations, human oversight in certain cases, simple revocation mechanisms, and clear accountability in case something goes wrong.
Banks are already accustomed to working with strict rules for payments, fraud, enhanced authentication, and transaction monitoring. The difference is that AI agents can operate in more dynamic environments. They can interpret instructions, query data, combine tools, and act on behalf of the user. That flexibility is useful, but it complicates control mechanisms.
At this point, financial innovation approaches a fundamental question: how is AI authorized to act? It’s not enough for the user to say “pay this” if the system can interpret, prioritize, or automate subsequent decisions. Trust will need to be built on policies, identity, verifiable records, and well-defined technical limits.
AI risk is also a data problem
The CSA and Anjuna report identify the leakage of sensitive data through AI interactions as the main security concern, cited by 61% of respondents. This is significant because it surpasses other more striking fears, such as direct attacks on the model or adversarial techniques.
In practice, the most immediate problem is not necessarily an “hacked” AI but an AI connected to too much data, with unclear permissions or insufficient visibility. An agent that summarizes files, consults customer information, interacts with internal systems, or helps resolve incidents can expose data if solid controls over what it can see, do, and record are not in place.
This concern aligns with another data point from the study: 20% of organizations acknowledge security incidents related to AI, while 21% are unsure if incidents have occurred. The latter figure may be the most worrisome; in a regulated sector, not knowing whether an incident has taken place can be as problematic as the incident itself, impeding investigation, correction, and demonstrating control to regulators and clients.
AI security risks in finance increasingly resemble cloud risks from a decade ago, but at a faster pace. First comes adoption. Then, visibility, identity, permissions, third-party integrations, and configuration issues emerge. Finally, governance frameworks are implemented. The key difference is that agents can act with much more autonomy than traditional applications.
Cloud, third parties, and human errors still weigh heavily
The report confirms that cloud services are now well-established in financial services. 98.3% of surveyed organizations use some form of cloud service, and one-third rely mainly or entirely on cloud-based architectures. This is important because enterprise AI increasingly depends on cloud infrastructure, distributed data, managed services, and models deployed in hybrid environments.
Senior management also seems to understand the link between cloud, security, and AI. 91% report having moderate or strong support from executive teams. This support is essential for funding projects, changing processes, and handling regulatory complexity. However, strong leadership support does not eliminate operational risks.
Main cloud security concerns remain very human: third-party risks, cited by 55%; misconfiguration, 52%; and human error, 27%. This clearly indicates that even with better tools, governance, integration, and operational failures continue to be the weak points.
In financial services, these risks are amplified due to the dependency on cloud providers, data platforms, external models, integrators, fintechs, APIs, and digital supply chains. An AI agent operating within such an environment needs controls not only over the model but over everything the model can access.
The next phase will focus on governance, not just adoption
The report concludes that AI in finance is entering a more mature phase. Competitive advantage will no longer be solely in deploying agents faster than others but in demonstrating that these agents operate within clear boundaries, with traceability, security, identity management, and continuous compliance.
This requires organizations to work on multiple fronts simultaneously. They must define what agents can exist, what data they can access, what decisions they can make, when human approval is needed, how results are audited, and how to respond to errors. Policies around third parties, cloud controls, identity management, and compliance frameworks will also need updating to ensure autonomy does not outpace supervision.
CSA’s message is particularly relevant in a trust-based sector. A bank can adopt AI quickly, but without clear mechanisms to control its agents, explain data usage, and decision processes, the promise of efficiency can turn into reputational and regulatory risks.
AI is already entering financial operations. The next step is not simply deploying more agents but building a control architecture to sustain them. Banks learned years ago that cloud required new rules. Now, they will have to learn the same with autonomous AI.
Frequently Asked Questions
What is agentic AI in financial services?
It is the use of AI systems capable of acting within financial processes, consulting information, using tools, automating tasks, and in some cases, making decisions with a certain level of autonomy.
How many financial entities are already using AI agents?
According to the Cloud Security Alliance and Anjuna report, 62% of surveyed financial organizations have already deployed AI agents.
What is the main security risk?
The primary concern is the leakage of sensitive data through AI interactions, cited by 61% of respondents.
What are autonomous payments with AI?
They are payments initiated or executed by AI agents on behalf of consumers. 85% of respondents believe they will occur, though most agree that new authorization models will be needed.