Yellow robotic arms near a conveyor belt

Agentic AI Trading: What Could Possibly Go Wrong?

Key Takeaways

AI Agents Introduce New Risks to Financial Markets
Autonomous trading could amplify cybersecurity threats, accelerate market disruptions and raise questions about accountability when things go wrong.

Old-School Safeguards Still Have Value
Diversifying financial relationships, maintaining independent records and preserving liquidity can help investors protect their wealth against emerging technological risks.

Keep Humans in Control of Financial Decisions
Until stronger safeguards are established, investors should exercise caution when allowing AI agents to execute transactions without human oversight.


Almost a decade ago, I penned an article for Inc. magazine advocating a low-tech approach to wealth management at a time when “more information about your finances is floating around the digital world – at precisely the same time the threat of cyber attacks is increasing.”

Prescient? Maybe. But I never envisioned that I’d one day read an announcement like the one Robinhood made earlier this year, introducing “Agentic Trading” and “Agentic Credit Card” accounts that enable AI agents to trade and make credit card purchases. “Bring your agents from anywhere and simply connect them to Robinhood’s AI-native Model Context Protocol (MCP) servers for fast, seamless integration into the Robinhood experience,” the firm wrote.

What could possibly go wrong? 

At financial industry conferences and regulator gatherings, warnings about the risks of AI agents – both granular and systemic – are invariably on the agenda. “Wall Street is racing to give artificial intelligence more control over investing,” said one report at the Securities Industry and Financial Markets Association’s recent annual meeting. “But the same technology that promises faster trades, cheaper advice and unprecedented access to financial markets could expose investors to a new generation of cyberattacks. … Protecting client assets and maintaining investor trust will become increasingly challenging as AI agents, tokenized securities and around-the-clock trading converge.”

We tell ourselves we’ve built layers of protection into the financial system. The reality looks more like 70 pieces of software layered on top of each other like ancient sediment. … In the new [Claude] Mythos world [of Agentic AI agents], the incumbent system has many thousands of vulnerabilities.

- Eric Peters, One River Asset Management Co-Founder and CIO 

What Could Go Wrong?

What are some of the new risks, or newly enhanced risks, created by AI agents?

To the extent personal AI agents created and deployed by individuals or institutions go rogue (as they already and famously have), it’s not hard to imagine the havoc they could create in individuals’ brokerage accounts. Or bank accounts. Or credit card accounts.

Then there’s compression risk. As the International Monetary Fund notes, AI is being increasingly embedded in the critical infrastructure of finance – trading, lending, supervisory protocols, operational workflows, cyber defenses and, yes, AI agents – making markets “faster and more tightly coupled” during periods of stress. Particularly as they evolve toward 24/7 “always open” status with less downtime to correct mistakes and implement circuit breakers.

A closely related version of compression risk is “herding,” often cited by former SEC Chairman Gary Gensler, which will only be enhanced by autonomous agents reacting in parallel to the same signals from a relatively narrow set of dominant AI systems. The analogy here is that of patrons in a movie theater all rushing for the exits when the fire alarm goes off.

Then add to that fiduciary risk. Institutional Investor magazine has written thought-provoking pieces suggesting that some asset managers may not have performed sufficient due diligence on the AI exposures of their portfolio holdings. They have called frontier AI safety “an unpriced, unmodeled risk sitting inside portfolios that touch AI.” I would say the same applies to AI agents acting on behalf of individuals. Are they mere counterparties? Transaction agents? Or fiduciaries? And who is ultimately liable for damages – the financial services firm? The developer of the frontier AI model used to create the agent? The individual?

Old-School Defenses for New Risks

The low-tech recommendations I offered up a decade ago predated and therefore don’t fully address the new AI agent risks above. But they do "anticipate the anticipation of trouble," as legendary market strategist Barton Biggs exhorted in his book Wealth, War and Wisdom. 

  • Diversify. Don't keep all your financial eggs in one basket. Intentionally disaggregate. If your investments are at a broker-dealer or a registered investment advisor, keep some of your savings – or at least your operating reserves – at an unaffiliated bank. Move funds between them when you need to. Sure, it's a little clunky, but it can create some separation between different buckets of wealth.
  • Make copies. Not so long ago – in fact, well within my professional lifetime – it was customary to receive physical delivery of pieces of paper that were either stock certificates or, more interestingly, bonds (weirdly fringed with detachable interest coupons). These documents were hard evidence that you actually owned a financial instrument. You could store them in your safe at home or in a safe deposit box at your local bank.

    Today, it is almost impossible to get your hands on a physical security. In the modern financial system, ownership records are almost universally in digital form. Financial data is stored and processed on servers or in "the cloud" or in wallets. But you can make and keep electronic or print copies of your statements. Print them out at home, or encrypt them and store them on flash drives. If you're an active investor, do this once a month.
  • Be real. Real assets, such as your home, farmland, gold coins or collectibles, can provide another layer of protection against risks in the digital financial system because they may be less vulnerable to unauthorized transactions by AI agents. If you own these assets, make sure you have solid documentation, including photographs and receipts, establishing your ownership.
  • Stay liquid. Remember Warren Buffett's adage: “To finish first, you first need to finish.” Make sure you have enough cash on hand to make it to the other side of a crisis. Lines of credit, including home equity or securities-backed loans, can provide additional flexibility, but remember that access to borrowed funds may be restricted when you need them most. And don't overlook the value of keeping some emergency cash secured in a safe place, independent of the financial system. Having readily available cash may also keep you from having to sell investments at inopportune times.

Now let me add one final recommendation for the AI age: Do not use autonomous AI agents to execute financial transactions on your behalf. At least not yet, not until guardrails are in place to protect you in an age when innovation-driven speed, automation and interconnectedness create new ways for things to break. Sometimes the best response to a new technological risk is an old-fashioned one: Keep a human in the loop.  

The information reflected on this page represents Baird experts’ opinions as of today and is subject to change. The information provided here has not taken into consideration the investment goals or needs of any specific investor and investors should not make any investment decisions based solely on this information. Past performance is not a guarantee of future results. All investments have some level of risk, and investors have different time horizons, goals and risk tolerances, so speak to your Baird Financial Advisor before taking action. The views and opinions expressed here are those of the speaker and do not necessarily reflect the views or positions of the firm.