Agentic AI

Scope, in the sources' words

• AIEOG AI Lexicon, "Agentic AI": "A category of AI systems capable of independently making decisions, interacting with their environment, and optimizing processes without direct human intervention." The Lexicon's footnote says the entry is "Adapted from" two cited publications.

• AIEOG AI Lexicon, "AI Agent": "A system that autonomously perceives its environment, decides what to do, and takes actions to achieve its goals."

• FINRA (2026 Annual Regulatory Oversight Report): "AI agents are systems or programs that are capable of autonomously performing and completing tasks on behalf of a user. An AI agent can interact within an environment, plan, make decisions and take action to achieve specific goals without predefined rules or logic programming."

• SR 26-2 (April 17, 2026), footnote 3: "Generative AI and agentic AI models are novel and rapidly evolving. As such, they are not within the scope of this guidance."

• The Lexicon's own limit: "The Lexicon is an optional tool and is not intended for use either in the legal interpretation of any regulations or regulatory oversight reports or supervisory statements of US financial regulators, international arrangements or agreements, or private contracts." This entry quotes it for vocabulary only.

• Which definition a particular team, contract, or regulator uses is a question for that team's compliance function or counsel.

What FINRA lists as risks of AI agents

FINRA's report lists "notable risks and challenges," including:

• "Autonomy: AI agents acting autonomously without human validation and approval."

• "Scope and Authority: Agents may act beyond the user's actual or intended scope and authority."

• "Auditability and Transparency: Complicated, multi-step agent reasoning tasks can make outcomes difficult to trace or explain, complicating auditability."

It adds that firms "may wish to consider" how "to monitor agent system access and data handling," "where to have 'human in the loop' agent oversight protocols or practices," "how to track agent actions and decisions," and "how to establish guardrails or control mechanisms to limit or restrict agent behaviors, actions or decisions."

Two kinds of agents in the news

Recent coverage of banks and AI agents involves agents that customers set up and give access to, and agents that institutions deploy themselves. The definitions above don't distinguish between them. Our post Who's Really on the Line? reports what banks have said about the first kind.

PayPath provides AI agents and a platform to orchestrate calls, texts, and emails for banks, credit unions, collectors, and other businesses. Its agents pause and bring in a person when they don't know how to complete an action or need clarification or confirmation, instead of guessing. See Human-in-the-Loop.

AI-agent operations questions

• Which of the definitions above does the team's vendor contract, policy, or risk framework use, and does anyone know?

• Where does the team draw the line between a tool that follows scripted rules and one that, in the sources' words, acts "without predefined rules or logic programming"?

• For each system that acts without a person's approval, who owns it, and what is logged?

• If the team's model risk framework follows SR 26-2, where does a system its footnote describes as outside the guidance get governed instead?