We Counted AI Hallucination Disclosures in Every 10-K. Pharma Outnumbered Software 29 to 10.
Thirty public companies told the Securities and Exchange Commission about "hallucinations" in their annual reports in 2022. ChatGPT was released in November of that year, near the end of it, so for most of 2022 the product that made the word famous did not exist. Thirty filings, in a year that mostly predates the thing everyone now means by the term. That number should feel wrong, and the…
In the first half of 2022, 30 companies disclosed hallucinations in their annual reports, despite the fact that the AI model ChatGPT was not released until November of that year. This discrepancy highlights the need to examine how corporate America disclosed AI risks in their filings. By searching SEC filings for the term "hallucinations" and counting them by year, a rising line of numbers emerged: 30 in 2022, 32 in 2023, 42 in 2024, 54 in 2025, and 166 so far in 2026.
However, this number is misleading, as many of the companies filing about hallucinations are actually pharmaceutical firms dealing with neurological symptoms, not AI risks. For instance, ACADIA Pharmaceuticals, a company specializing in treating hallucinations associated with Parkinson's disease, frequently mentioned hallucinations in its 2026 10-K filing.
This demonstrates that the original count was capturing the base rate of a neurological symptom in biotech companies' annual reports, rather than AI risks. To correct this, the search was narrowed to include "hallucinations" and "artificial intelligence". This adjustment brought the 2026 count down to 159, but the pharmaceutical industry still dominated the list of filers, with 29 filings compared to 10 for software companies.
The error lies in the fact that mentioning artificial intelligence in a 10-K filing has become common practice, so requiring the term's presence is no longer an effective filter for measuring AI-related disclosures. The three different methods of counting filers yield vastly different results - from 2 to 159 - and the discrepancy between them is not noise, but rather a manifestation of the measurement error.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — it may contain errors, so check the original before relying on it.
