SEC Chair Warns of Risks in AI-Related Fraud
Commentary by Steven Lofchie
SEC Chair Gary Gensler warned that artificial intelligence ("AI") introduces new avenues for fraud and market manipulation under securities law.
In an "Office Hours" video, Chair Gensler described three categories of potential harm related to the use of AI in the financial sector: "programmable harm," "predictable harm," and "unpredictable harm".
He explained that programmable harm occurs when algorithms are intentionally designed to defraud or manipulate the public. He said, "if someone uses an algorithm and is optimizing to manipulate or defraud the public, that's just fraud."
"Predictable Harm," he said, involves situations where individuals or firms recklessly disregard foreseeable risks when deploying AI models. He said the test is whether those using these models acted reasonably. He argued that "investor protection requires that the humans who deploy the model put in place appropriate guardrails." He asserted that financial entities must implement safeguards to prevent reckless behavior and ensure that AI models do not engage in practices like front running or spoofing.
Addressing the third category, "Unpredictable Harm," Chair Gensler acknowledged the challenges in holding firms accountable for unforeseen consequences of self-learning AI systems which may evolve unpredictably or as "hallucinations." Nevertheless, he said that firms must take responsibility for the acts of their AI systems. Citing former SEC Chair Joseph Kennedy's commitment to fighting fraud, Chair Gensler argued that the same resolve applies today, warning that those deploying AI without proper controls will face consequences under securities law.
Commentary
Chair Gensler divides AI risks into three types, and then says, never mind, the SEC is going to treat them all the same. It would have been helpful had he considered how the differences might manifest themselves in particular situations and how the SEC would view those differences.
The irony is that the regulators tout the uses that they can make of AI, even though AI is not fully predictable. Yet, the regulators treat use of AI by firms as creating not merely strict liability, but also an obligation to know how AI produced each result, even though the regulators know that AI results in some unpredictability.