_{Challenges of Agentic AI}

_{The rise of Agentic AI marks a major shift in artificial intelligence, enabling systems to make decisions, take actions, and operate autonomously within financial services and beyond. While this evolution brings significant opportunities, it also introduces technical, ethical, regulatory, and operational challenges that must be addressed for effective and responsible deployment. Financial institutions, regulators, and technology leaders must grapple with these complexities to ensure Agentic AI operates safely, ethically, and within legal frameworks.}

Regulatory and Compliance Uncertainty

The deployment of Agentic AI in financial services raises complex legal and regulatory questions that have yet to be addressed. Unlike traditional financial decision making, which follows clear compliance frameworks, AI driven financial models could create regulatory grey areas, leaving institutions uncertain about liability, oversight, and governance.

A key challenge is determining accountability when AI systems make financial decisions. If an autonomous AI agent executes a poor investment, mismanages risk, or fails to detect fraud, where does liability fall? Should responsibility lie with the AI developer, the financial institution deploying the AI, or the client relying on AI driven advice? These unanswered questions present significant compliance risks for financial firms.

If an AI Agent makes a bad call – who is responsible?

Regulatory bodies have yet to fully align global AI governance frameworks, making compliance particularly difficult for multinational financial institutions. The EU AI Act, U.S. AI regulations, and regional compliance standards introduce further complexity, forcing financial institutions to navigate a fragmented regulatory environment. To address these challenges, institutions must establish robust internal AI governance frameworks, collaborate with regulators to define industry best practices, and adopt "compliance by design" AI models that integrate legal and ethical considerations from the outset.

As has been noted, the adoption of AI has suddenly surged in recent years and is evolving at an extremely fast pace. As society adapts to our future technology state, politicians are closely reviewing how the world is changing. In the UK in February this year the Treasury Committee has launched a call for evidence of the impact of AI on banking, pensions and financial services. Political desire to adopt AI for beneficial purposes is tempered by inspection and consideration of whether and how failures could occur, whether at the individual or systemic level in the financial system.

Operational and Integration Challenges

Beyond regulatory concerns, financial institutions face significant operational hurdles when integrating Agentic AI into existing infrastructures. Many financial firms rely on legacy banking systems, which were not designed to support fully autonomous AI agents. The incompatibility between outdated IT infrastructure and modern AI models can hinder efficiency and limit the ability to scale AI driven automation. Data fragmentation further complicates AI adoption. AI models thrive on high-quality, real-time financial data to function effectively, yet many institutions operate in siloed, fragmented environments, where critical data is spread across multiple systems and departments.

Without seamless data integration, AI systems may lack the comprehensive insights needed for effective decision making. Additionally, the shift toward AI driven financial automation introduces organisational resistance, as employees worry about job displacement and struggle to adapt to new AI enabled workflows.

Successful AI adoption requires careful change management, with

institutions investing in AI native infrastructure, breaking down data silos, and upskilling employees to ensure a smooth transition.

While Agentic AI offers transformative potential in financial services, its adoption comes with significant challenges that must be proactively addressed. Issues related to explainability, bias, cybersecurity, regulatory uncertainty, and operational complexity will define the future of AI deployment in the financial sector. Financial institutions must take a strategic, well-governed approach to AI adoption - one that balances innovation with oversight, autonomy with accountability, and efficiency with security.

Embedding AI transparency, ethical safeguards, and compliance measures into AI driven systems from the outset will be critical in unlocking the full potential of Agentic AI while maintaining trust, reliability, and regulatory alignment. The future of AI in finance is no longer a question of whether institutions should adopt Agentic AI - it is a question of how quickly they can build the right frameworks to support its safe and responsible deployment.

The Psychological and Behavioural Impact of DIFM

While the benefits of DIFM are undeniable, its widespread adoption raises important questions about human behaviour, trust, and control. As AI takes on more decision making responsibilities, users may become increasingly detached from direct, manual problem solving and financial literacy, potentially reducing their ability to make informed decisions independently.

The convenience of the “Do It For Me” economy could lead to an over-reliance on AI, where individuals no longer actively engage with important aspects of their financial position, health, and management of personal affairs. Trust in AI driven systems also becomes a critical issue. Consumers must be confident that these systems are acting in their best interest, particularly in high-stakes areas like finance and healthcare.

Transparency, accountability, and explainability will be crucial in maintaining trust, ensuring that AI driven decisions are both understandable and ethically aligned. Additionally, as AI becomes more hyper-personalised, concerns around data privacy, security, and ethical use will become more pronounced.

Consumers expect AI to anticipate their needs, but there is a fine line between personalisation and intrusive decision making. Businesses will need to balance automation and human oversight, ensuring AI driven services remain responsible, transparent, and aligned with user expectations.

One of the most pressing concerns surrounding Agentic AI is its lack of explainability, often referred to as the "black box" problem. Unlike traditional financial models that follow predefined rules, these systems learn from vast datasets and refine their decision making processes independently. The result is a model that can make autonomous choices, yet the logic behind those decisions may not be readily understood by humans. This lack of transparency creates challenges for regulators, financial professionals, and customers, as it becomes increasingly difficult to audit, challenge, or fully trust AI driven financial decisions.

Regulatory bodies demand clear decision making rationales, particularly in areas like lending, investment management, and risk assessment, where opaque AI driven processes could lead to legal and financial risks. Compliance teams may struggle to explain AI generated recommendations, raising concerns about regulatory violations and heightening reputational risks for financial firms.

Additionally, customers are far less likely to trust AI driven services if they cannot understand the reasoning behind key financial decisions that impact their wealth,

creditworthiness, or risk exposure. Addressing this challenge requires the development of explainable AI (XAI) models that enhance interpretability. AI systems must generate audit trails that log decision making processes, ensuring financial institutions can review, justify, and validate AI driven actions.

Additionally, human-in-the-loop mechanisms may be necessary for high-stakes decisions, ensuring a balance between AI autonomy and human oversight.

AI must be designed so it’s decision making is explainable, and with humans-in-the-loop for oversight.

Agentic AI systems rely on historical financial data to make decisions, but if these datasets contain biases, AI can replicate and even amplify existing discrimination.

This is particularly concerning in sectors like credit approvals, loan disbursements, wealth management, and fraud detection, where biased algorithms could disproportionately disadvantage certain demographic groups. AI driven lending models, for example, may inadvertently discriminate against minority communities if trained on biased historical data, reinforcing systemic inequalities in access to credit. Fraud detection systems may unfairly flag transactions from underrepresented customer groups, leading to higher rates of false positives and financial exclusion.

In investment management, AI driven strategies may disproportionately favour larger institutions, making it harder for smaller investors to compete. Regulatory bodies such as the FCA and SEC are increasing scrutiny on AI driven financial decision making, demanding greater oversight to prevent bias and discrimination. Financial institutions must proactively address these risks by training AI models on diverse and representative datasets, conducting bias audits, and establishing AI ethics committees to oversee decision making processes.

Ethical AI frameworks should be embedded into AI governance structures to ensure fairness and compliance.

As Agentic AI assumes greater autonomy, it also introduces new cybersecurity vulnerabilities that malicious actors may exploit. The financial sector, a prime target for cybercriminals, faces heightened risks as AI driven systems take on more decision making power in trading, fraud detection, and asset management.

AI models are susceptible to manipulation through adversarial attacks, where bad actors introduce deceptive inputs to mislead AI systems into making incorrect financial decisions. Fraudsters can also leverage AI to automate sophisticated financial crimes, such as deepfake-powered scams or AI driven market manipulation. In the trading sector, fully autonomous Agentic AI models could be exploited to trigger flash crashes or manipulate stock prices through coordinated trading attacks.

Prompt Injection: Criminals input deceptive prompts to influence AI-driven financial decisions, bypassing safeguards to manipulate trades, approvals, or fraud detection for illicit gain.

The severity of AI driven cyber threats is drawing increased regulatory attention, with financial regulators urging institutions to strengthen AI security protocols. To mitigate risks, firms must implement adversarial defence mechanisms, develop AI driven cybersecurity tools to detect and neutralise threats in real time, and explore blockchain-based AI governance to prevent unauthorised tampering with AI models. The integration of real-time anomaly detection and proactive cybersecurity measures will be critical in safeguarding AI driven financial systems.