- Agentic AI offers powerful automation and personalization for finance
- IBM warns of risks like goal misalignment and
data privacy breaches
- Systems act autonomously, creating governance
and safety challenges
- Financial firms are adopting agentic AI for KYC,
AML, and risk
- Infosys and PwC report significant efficiency
gains with AI agents
- Metzler Bank partners with AI firm Unique for
back-office automation
- AI agents can misuse APIs or leak sensitive data
unintentionally
- Adoption is expected to grow by 600% by 2026
- Regulatory compliance may be at risk if
oversight fails
- IBM urges proactive governance and ethical AI
deployment
Agentic artificial intelligence is being hailed as the next major force in financial services, promising hyper-personalization, frictionless compliance, and back-office optimization.
However, according to a new IBM report, the same characteristics that make agentic AI so powerful also make it inherently risky.
Defined as AI systems with goal-directed behavior and autonomy, agentic AI can plan, execute, and adapt actions with minimal human intervention.
These agents are being trained to interpret complex data, create strategic responses, and carry out tasks across customer-facing and operational domains. But that independence also opens the door to unforeseen outcomes, privacy violations, and compliance breakdowns.
In financial services, use cases for agentic AI are expanding rapidly. IBM outlines its transformative potential across areas such as onboarding, fraud detection, risk assessment, and KYC.
In one example, an agentic AI system handles the end-to-end account setup process, delegating tasks like sanctions screening and document verification to specialized agents. If anomalies arise, the system can escalate the case to a human for final approval.
Customer personalization is another high-impact frontier. Agentic systems are being deployed to provide real-time product recommendations, dynamic pricing, and tailored robo-advice – drastically enhancing client engagement while reducing operational strain.
The ability to orchestrate complex workflows without constant oversight is what makes these systems so appealing to firms looking to modernize.
But there’s a catch. As IBM points out, these systems also carry significant risks. Goal misalignment is one of the most concerning. AI agents may pursue their assigned objectives too rigidly or interpret goals in ways that contradict company policy or ethics.
Their growing autonomy means they may combine APIs or tools in unanticipated ways, creating vulnerabilities or taking unauthorized actions that bypass human oversight.
Privacy is also a growing concern. Agentic systems access, process, and sometimes store sensitive data.
In doing so, they may unintentionally leak confidential information or breach compliance requirements, particularly if persistent memory or unrestricted integrations are involved.
The danger is not just that they might fail—it’s that they might succeed in ways no one intended.
Despite these risks, adoption is gaining momentum. Infosys reported dramatic efficiency gains in software development using AI agents, including up to 90% improvements in database code generation.
PwC has deployed agentic systems to automate tax workflows involving notoriously complex forms. Meanwhile, German bank Metzler has partnered with Swiss startup Unique to deploy off-the-shelf agentic AI use cases in its back-office operations.
Investors are also taking note. Unique raised $30 million in Series A funding earlier this year to expand globally, particularly into the US market.
Yet as the technology spreads, so too must caution. IBM’s report makes it clear: the financial sector cannot afford to treat agentic AI as just another productivity tool. Without rigorous oversight, ethical design, and legal accountability frameworks, the very features that make agentic AI valuable could also make it catastrophic.