With most calendar-year end companies having filed their proxy with the SEC for this proxy season, I thought it would be a good time to reflect upon the risks of relying on AI to help draft disclosure since the continuing improvement of AI models (known as “LLMs”) has reduced – but certainly not eliminated – the risks of hallucinations.
I’ll cover ways that AI can be helpful during the drafting process soon enough – but I thought I’d start with a realistic example to illustrate the stark gap between what AI likely would draft compared to what a competent disclosure lawyer would. We should remain vigilant and not overly rely on AI and the case study below should help drive that point home:
Here’s what AI wrote as a sample paragraph for the CD&A:
“In fiscal 2025, the Compensation Committee approved an increase in CEO compensation reflecting strong company performance. Revenue and Adjusted EBITDA grew by 12% and 15%, respectively, and total shareholder return exceeded the median of the peer group. The Committee also updated the long-term incentive program to include additional performance metrics aligned with shareholder value creation. These actions demonstrate the Company’s commitment to pay-for-performance alignment.”
At first glance, you may be taken aback because it does sound like our brand of legalese. But when you take a second look, you can see how generic it is. It reads like you would think someone who is poking fun would write it. As you can tell, AI is good at producing statistically “normal” disclosure and mimicking common phrasing.
But what it winds up with is strictly boilerplate. It doesn’t know your company’s facts. And it doesn’t apply judgment about materiality or investor and proxy advisor perception. In fact, it provides a good example of the type of vague proxy disclosure that proxy advisors and investors abhor – and that plaintiffs’ lawyers could be drawn to when considering which disclosures should be the target of a lawsuit.
In comparison, this is how a human disclosure drafter would likely write up that CD&A paragraph:
“In fiscal 2025, the Compensation Committee increased the CEO’s total compensation by 18%, reflecting performance that exceeded pre-established targets across key financial metrics. Revenue increased 12% and Adjusted EBITDA grew 15%, both above target levels set at the beginning of the year.
Total shareholder return ranked at the 65th percentile relative to the Company’s peer group, reinforcing alignment between realized pay outcomes and shareholder experience.
In determining compensation, the Committee also considered the CEO’s role in advancing the Company’s strategic priorities, including the expansion of higher-margin product lines and operational efficiency initiatives.
The Committee approved the addition of a relative TSR modifier to the long-term incentive program beginning in fiscal 2025 to further strengthen the alignment between executive compensation and shareholder returns.”
What changed? A number of important things:
- There is specificity over generality, such as tweaking the phrase “strong performance” to “exceeded pre-established targets.”
- There is a clear pay-for-performance linkage, with the human drafter’s disclosure providing an explicit connection between metrics, targets, outcomes and the resulting pay.
- There is transparency about the compensation committee’s use of judgment by covering strategic considerations and qualitative factors.
- There is a calibration in tone to remove fluffy or defensive language to make it more measured with factual phrasing.
- A change to LTI design is highlighted, likely to illustrate that the board was responsive to concerns expressed during shareholder engagement or a low say-on-pay vote received the prior year.
Authored by
Broc Romanek