Nadia ANTONIN
Since 2022, the debates on artificial intelligence (AI) have mainly focused on language models and conversational assistants (chatbots), such as ChatGPT (OpenAI), Copilot (Microsoft), Gemini (Google), Claude (Anthropic) or Chat (Mistral). In this regard, there are different formats of conversational robots, including one based on generative AI using machine learning machine learning and natural language processing.
After chatbots and generative AI, a new generation of AI systems is emerging: agentic AI, which is already being used in many sectors. In the financial sector, this technological development opens up significant opportunities, but also raises major challenges.
Agentic : a: conceptual analysis
“The era of agentic AI has already arrived. Agents are being deployed on a large scale in the economy to perform all kinds of tasks,” said Sinan Aral, Professor of Management, Computer Science and Marketing at MIT Sloan.
A report by the MIT Sloan Management Review and the Boston Consulting Group in November 2025 reveals that 76% of executives now consider agent AIs more as colleagues than as tools. Moreover, in less than two years, 35% of companies have already adopted agentic AI and a further 44% plan to deploy it very soon.
As the National Institute for Research in Digital Science and Technology (INRIA), agentic AI represents less of a fundamental scientific breakthrough than a functional and systemic breakthrough. For the AI and Digital Council, Digital Council, agentic AI does not constitute a single technological breakthrough but rather corresponds to the combination of technological accelerations.
The Organisation for Economic Co-operation and Development (OECD) points out that the concepts of “AI agents” and “agentic AI” are sometimes used interchangeably, which leads to some confusion and a lack of conceptual However, closer examination reveals important distinctions between these two concepts (“The agentic AI landscape and its conceptual foundations”, February 2026).
For the OECD, “Agentic AI refers to systems composed of multiple coordinated AI agents that are coordinated, capable of breaking down tasks, collaborating and pursuing complex goals autonomously over long periods of time. These systems are designed to operate in more open and less predictable physical or virtual environments that are more open and less predictable, and to operate with minimal human supervision “.
Unlike generative AI, which is limited to producing content, agentic AI, an autonomous autonomous AI system, is capable of acting independently to achieve predetermined objectives. It is delegated a task to perform. It consists of AI agents, machine learning models capable of imitating human human decision-making to solve problems in real time.
The main challenges of agentic AI in the financial sector
Agentic AI can automate complex processes to improve productivity while reducing operational costs: market analysis and business intelligence, detection of fraud and suspicious transactions, automation of compliance processes, etc.
In addition, agentic AI makes it possible to accelerate decision-making thanks to better use of data: improving the relevance of the data used and consolidation of previously dispersed data.
Finally, thanks to agentic AI, AI, bankers, insurers and asset managers can offer personalised financial services.
Some examples of the use of agentic AI in the banking and financial sector
This new AI system is booming in US banks. According to a survey by Ernst and Young and MIT, by 2025, 70% of banking sector executives say their banks are already using agentic AI. In the field, the applications in the banking sector are multiplying. For example, JPMorgan Chase is exploring the use of agentic AI to detect fraud, provide personalised financial advice, automate loan approval processes processes and process legal documents.
In France, Société Générale, through its new entity SocGen AI, is transforming its digital transformation by focusing on agentic AI, which it sees as the next major step beyond traditional generative AI.
The BPCE Group is relying on AI to automate more processes, while keeping humans at the heart of strategic decisions. Tests are being carried out on different business areas: customer relations, middle and back offices, controls.
The main challenges related to agentic AI
One of the biggest challenges concerns the question of liability in the event of a failure of an autonomous system. Hence the need to maintain human oversight.
In addition, with agentic AI, AI, we face the risks of regulatory non-compliance (GDPR, European AI Act, anti-money laundering, etc.).
Regarding cybersecurity, while agentic AI can be put to use (detection of fraud, in particular), it nevertheless creates new threats: attacks by injection of generative instructions (prompts), leakage of sensitive data, leaks, etc. A Senate report of 28 April 2026 highlights that agentic AI accelerates attack capabilities much faster than the installation of cyber defence tools.
In addition, the proliferation of agentic AI leads to new systemic risks such as the amplification of market reactions, the increase in financial contagion, etc.
Finally, among the ethical challenges of agentic AI, its deployment poses risks of algorithmic bias bias and breaches of personal data privacy.
As conclusion, agentic AI now heralds profound changes within the finance sector. finance sector. It is gaining ground by automating complex processes, by establishing itself as the new driver of customer relations and improving risk management. However, its deployment raises considerable challenges in terms of governance, regulatory compliance, cybersecurity and systemic risk management. To succeed in this new revolution of artificial intelligence, it will be necessary to combine agent autonomy, human supervision and regulatory compliance. As Bpifrance points out, agentic AI, AI, however powerful it may be, has its limitations and risks. “The balance between autonomy, supervision and human responsibility remains essential to ensure the reliability and the ethics of the decisions made.”