The emergence of agentic artificial intelligence is turning AI into a powerful force multiplier in the production and distribution of prospectuses and other regulatory communications, but in so doing, it also underscores the need for human judgment and oversight.
Agentic AI differs from earlier iterations of AI technology in one important way: Previous AI models followed instructions. Agentic AI achieves goals. When an agentic AI model is assigned a task, it works autonomously to figure out the best way to accomplish that goal. That ability sets agentic AI apart from all other forms of AI, and it dramatically increases its practical value to companies, who are already rolling out real-life use cases throughout the entire life cycle of regulatory document production and distribution.
The rapid proliferation of agentic AI applications raises legitimate concerns about risk, and about the potential impact the technology will have on jobs in regulatory communications and across the labor market. However, in my opinion, the fact that agentic AI functions as an autonomous problem-solver will actually increase the need for human involvement.
No company, especially a company operating in a regulated environment, can afford to have a technology executing core business functions without human controls. Companies need to understand where their AI applications are getting information and how they are reaching conclusions. Companies need to assess the outcomes generated by AI models and determine if they are reliable, where they are appropriate for use, and how they should be implemented.
For these reasons, I believe most agentic AI applications won’t function independently within organizations but will instead operate in close coordination with humans. I also believe those partnerships have the potential to exponentially enhance the productivity of individual employees and to streamline regulatory communications into a process that is faster, less risky and much cheaper.
Agentic AI Use-Cases Targeting Regulatory Communications
Some of the most effective and durable agentic AI use-cases rolled out by fund companies so far are aimed at reporting and communications.
For example, companies creating regulatory communications documents draw content from many different places, including internal data, data from administrators like State Street, and other disparate sources. One of the biggest challenges for regulatory communications teams is making sure the right pieces of data and the right content are making it into the regulatory document. Today, companies are using agentic AI to validate documents by checking to see if any individual pieces are missing.
Regulatory communications teams are also using agentic AI to validate that the data that has been included in documents is correct. Asset managers produce a host of regulatory documents including prospectuses, annual reports, market documents and many others. Because many of these documents contain overlapping content, it is up to the team to ensure that every piece of data in every footnote across all those documents is the same. Agentic AI applications are acting as a backup set of eyes, analyzing documents after teams have completed their checks to ensure that nothing was missed and data is consistent across all documents.
Agentic AI is also being used by key external partners like Broadridge to streamline processes and reduce the amount of time companies have to spend on distribution. Every day, Broadridge ingests a huge number of documents from our clients that we share directly with investors. It’s up to us to ensure we’ve received the right documents, and that the right documents are shared with the right shareholders. Today, we are running an AI-automated validation application that acts as a safety net, flagging any anomalies far enough upstream in the process that our team has time to remediate before problems arise. We are also using AI to help streamline content tagging and other processes.
Calibrating Agentic AI Autonomy
As companies implement AI solutions in regulatory communications and elsewhere, they will have to decide where it’s appropriate to deploy agentic AI and how much autonomy to give agentic AI applications. This decision is part of a broader discussion about AI risk. At every level, the question companies should be asking is not, “What can AI do for us?” but rather, “What can AI do for us without creating an unacceptable risk profile?”
In that vein, decisions about how much autonomy companies give agentic AI applications will vary from firm to firm, and within organizations, from function to function. For example, risk tolerance in a tax function will be much lower than in non-mission-critical functions, meaning companies will likely allow lower levels of autonomy.
I believe companies should be inherently conservative when it comes to any client-facing content. For best practice, companies should never send any content to clients that has not been vetted by a human. For now at least, that rule should be a strong guardrail that restricts AI autonomy in client-facing communications.
Embedding AI Governance in Workflows
As AI models continue to evolve and workers become more experienced and comfortable with the technology, it’s almost inevitable that companies will become more expansive in their policies about agentic AI use and autonomy. For that reason, it is incumbent on companies to establish effective AI governance.
AI governance cannot exist at only a policy level. Governance policies that function as an overlay 10,000 feet above day-to-day operations won’t be enough to mitigate the real risks surrounding AI adoption. In regulatory communications, AI applications that are 95% accurate won’t cut it. Companies need 100%, every time. For that reason, strong governance rules must be embedded in workflows to ensure that the data going into AI applications is accurate and that decisions and output flowing from AI applications are sound.
What should effective AI governance entail? At a minimum, the company’s policies should ensure a traceable lineage of all AI output, and an immutable record of AI processes that can be audited at any time.
In all AI applications, companies must be able to answer core questions like: What was the source of the data the AI model used? What did the AI model do with that data? Did anything change from the original source to the final output? In governance terms, companies answer all those questions by establishing traceable content lineage and provenance.
The record of how an AI model generated its output must be both transparent and immutable, especially in highly regulated industries. Audits can occur at any time, even long after the fact. Companies need to be able to go back in time to document exactly what data was fed to an AI model, how the AI model reasoned, and how it came up with final output. That’s not as easy as it might sound. For example, AI models are changing all the time. By the time a company is audited, the model it used in a specific task might have been retired by OpenAI or Anthropic in favor of a new version. Companies need enough documentation to go back to the original AI model used to show how it came up with its result.
Of course, since we are living in the age of AI transformation, companies are using agentic AI to tackle the issue of AI compliance. AI-driven compliance agents are being deployed at every stage of the workflow to ensure visibility, auditability and repeatability in any and all AI applications running within the organization.
In regulatory communications, these agents will allow companies to safely apply more AI applications and to give agentic AI more autonomy to complete tasks. The proliferation of increasingly independent agentic AI agents will, in turn, will require more human judgment and oversight. The resulting partnerships between AI and human workers will vastly increase the productivity of regulatory communications teams, resulting in faster turnarounds, lower costs and reduced risk.
