top of page

Hudson Labs Company Due Diligence - The Highlights View

The “Highlights” view shows filings where we’ve identified one or more red flags. Refer to our FAQ for a list of the securities filings that we cover. The default version of the ‘Highlights’ view shows only medium and high severity red flags that are new to the company. We define new vs. old red flags below.

Highlights section of Bedrock timeline

About red flags

A red flag is a sentence or paragraph that has an empirical relationship with outcomes related to earnings management or fraud.

We use language models to find textual disclosure that has historically been predictive of fraud, malfeasance or earnings management. We use SEC enforcement actions and settled class action lawsuits, as well as other labels, for model training and testing. In other words, we automatically identify the in-text information that would be highlighted as important by a forensic analyst.

Learn more about our approach to language modelling and NLP research on the “AI in Finance” section of our blog.

Old vs. new red flags

By default we show “new” red flags in the “Highlights” view. A new red flag is one that has never appeared before (since 2019) in any securities filing we’ve previously processed. We test newness on a sentence basis, therefore, one new word will cause the disclosure to be tagged as “new”. Red flags are new unless you see an “Old” tag. The “Old” tag will tell you when the sentence was first used/disclosed.

old red flag with date first filed


After extraction, red flags are categorized. Learn about our categories here. To view red flags in a specific category across time, use the “Categories” drop down.

dropdown for red flag categories

No keyword reliance

Because we use language models, red flags are identified regardless of how the company chooses to disclose it. Whether a company discloses an “SEC subpoena”, or a “government request for documents”, the model is assessing that as the same information with the same level of severity.


Red flags are assigned a severity of high (red flag), medium (orange flag) or low (yellow flag). The severity is based on a) how predictive the information is of fraud and b) the model’s confidence.

Severity of red flag indicated and marked by colored flag

When the model is not confident about relevance i.e. whether the disclosure is in fact related to a high risk outcome, it is more likely to make mistakes. This could be due to unusual wording. Lower confidence red flags (yellow flags) generally include more false positives than higher severity categories.


bottom of page