Are Form 10-Ks Becoming Less Specific?

This new paper from MKAI – “The First Annual Reports of the LLM Era” – finds that disclosure language in large-company 10-Ks became less specific and more hedged at a faster rate during 2024 than in prior years. While this timing overlaps with the rise of enterprise LLMs, the study carefully avoids claiming causation – but instead frames the issue as a potential corporate governance and disclosure-controls challenge worthy of further monitoring:

1. Study Scope & Purpose: The paper analyzed 150 SEC 10-K filings from 50 large U.S. public companies across FY2019, FY2022, and FY2024 to measure changes in disclosure language. It examined six dimensions of “prose drift,” including increased hedging, loss of specificity, generic framing and sentence inflation. The study is descriptive and doesn’t claim that AI caused the observed changes.
   
2. Disclosure Language Drift Accelerated After 2022: On the five core editorial dimensions (excluding new-section effects), the annual rate of language drift increased from 0.71 to 0.88 points per company per year, a 24.5% increase in the FY2022–FY2024 period compared with FY2019 – FY2022.
   
3. The Nature of Drift Changed: Earlier changes were dominated by sentence inflation (longer passages without materially more information). In the later period, drift became more dispersed across several categories, including hedge proliferation, framing drift and loss of named specifics, making disclosure changes harder to detect through word-count analysis alone.
   
4. Sector Results Were Uneven: Technology, industrials and healthcare companies showed significant acceleration in drift during the later period, while financial services firms showed less drift and energy companies remained relatively stable. The paper cautions that sector subgroup sizes are small and findings should be viewed as indicative.
   
5. Timing Coincides with Enterprise LLM Adoption: The study notes that major enterprise-grade AI writing tools became widely available in 2023, making FY2024 the first filing cycle likely drafted after such tools became broadly accessible. However, the paper emphasizes that timing alone does not establish LLM use or causation.
   
6. No Evidence That AI Drafted Any Particular Filing: The paper explicitly states that many factors can change corporate disclosure language, including legal review, personnel changes, regulatory developments, standardization efforts and governance choices. The research documents patterns but does not attribute them to AI.
   
7. Some Companies Became More Specific, Not Less: Several companies, including NVIDIA and ExxonMobil, either maintained or improved disclosure specificity over time. Seven companies scored very low on drift, suggesting that higher-quality disclosure remained achievable under the same market and technological conditions.
   
8. Governance Implication: The paper argues that boards and disclosure committees may need to monitor not only disclosure length but also more subtle shifts in specificity, tone and informational sharpness. The concern is not necessarily AI use itself, but whether existing review processes can detect gradual deterioration in disclosure quality.