Recently, The New York Times reported that autonomous AI agents developed by OpenAI probed several U.S. government websites, including the SEC’s, without the company’s knowledge, and posted SEC information on an online forum. The SEC has said no nonpublic data was released. It is the latest in a string of incidents in which AI agents acted outside the boundaries their developers set.

Why This Matters for Your Firm

If AI agents can probe a federal regulator’s website, they can probe yours. An agent can scan for exposed systems, misconfigurations and unpatched vulnerabilities at a speed and scale that previously required a skilled team.

Practitioners pointed to a familiar weak point: the connection between public-facing data and the internal systems that feed it. Investor portals, websites and client reporting tools that draw on internal databases can become entry points if those connections are not segmented and access-controlled.

The good news is that the defenses are largely the basics, applied consistently. Access management, patching, segmentation and monitoring matter more, not less, when attackers are automated.

FINRA’s Tighter Login Controls: Stronger, but Plan Ahead

Much of the filing data the SEC relies on is collected through FINRA systems. FINRA has significantly strengthened credentialing for its Gateway, and even a single filing now requires a more rigorous multi-factor authentication process.

That is a meaningful control, but it also means more ways for access to fail. Firms should expect login and credentialing issues and should not discover them on the day a filing is due.

The Agents Inside Your Firm

The incident also raises a question closer to home: what are your firm’s own AI agents allowed to do? Many firms are adopting tools that can browse the web, read email, query internal systems and act on a user’s behalf. Even sophisticated developers cannot always predict how an agent will behave once it has access.

Firms using agentic tools should be able to answer a few basic questions. Which tools are in use, including those adopted by individual employees? What systems and data can each one reach? Which actions require human approval? Are agent activities logged and reviewable? And has the vendor been diligenced on how it constrains agent behavior and protects client data?

Regulators are already asking advisers to describe their AI use, supervision, data security and incident history. A firm that cannot inventory its AI tools will have difficulty demonstrating that its policies are reasonably designed under Rule 206(4)-7 or that customer information is protected under Regulation S-P.

An Unsettled Policy Landscape

The policy response to AI safety remains divided. This week, AI industry leaders signed a voluntary White House agreement on AI safety that the President described as morally rather than legally binding. Others, including Anthropic’s CEO and Rep. Maxine Waters, have called for firmer limits on the most advanced models.

For firms, the takeaway is practical: do not wait for new AI-specific rules. The Compliance Rule, Regulation S-P, the books and records rules and your fiduciary duty already apply to how you use and secure AI.

Five Steps Firms Can Take to Protect Themselves Now

• Inventory your AI tools. Identify every AI tool and agent in use, including tools employees adopted on their own, and what each can access.

• Limit agent permissions. Apply least-privilege access, require human approval for consequential actions and log agent activity.

• Update your incident response plan. Add AI scenarios, such as a compromised or misbehaving agent, to your next tabletop exercise.

• Diligence AI vendors. Assess how vendors constrain agent behavior, handle your data and notify you of incidents, consistent with Regulation S-P service provider requirements.

• Harden public-facing systems. Confirm websites, investor portals and reporting tools are segmented from internal systems and access-controlled.

Making Governance Operational

Silver’s AI Governance Program, powered by Clarier, helps firms discover AI tools in use, document AI use policies, assess vendor risk and report to leadership and boards. Our cybersecurity team also provides risk assessments, penetration testing, incident response planning and tabletop exercises tailored to investment managers and their portfolio companies.

To discuss your firm’s AI and cybersecurity readiness, contact Mike Regan, Director of Cybersecurity Compliance, at [email protected].