Early warning system: An experiment in tracking nonprofit closures and layoffs
Nonprofit closures and layoffs are happening faster than data can capture. Candid is using AI to track distress signals in real time—here’s what the early findings reveal.
Recent research has highlighted that the social sector is in a state of distress. Funding cuts and political pressure have resulted in financial instability for many nonprofits, leading to difficult choices around winding down programs, roles, and sometimes entire organizations. The impact of these decisions on individuals and communities are instantaneous: A food pantry closes its doors, an afterschool program gets canceled, an employee loses their job. But broader knowledge about these distress signals (i.e., nonprofit closures and layoffs) only surface months or years later.
That’s why we’re experimenting with a new AI-based tool that reads the news daily, picks out stories about nonprofit closures or layoffs, and captures the details we’d otherwise miss. Here’s how it works and what we’re seeing so far.
The lack of timely information on nonprofit closures and layoffs
It is incredibly difficult to access consistent data about nonprofit closures and layoffs. Technically, nonprofits are required to file a notice of termination with the IRS within 5.5 months after they close. However, that can be hard to enforce once an organization no longer exists.
Additionally, nonprofits, especially small ones, may temporarily disappear from the IRS Business Master File due to lack of up-to-date paperwork, so absence from this list may not necessarily imply a permanent closure. Similarly, while the WARN database tracks mass layoffs for large organizations (typically a layoff of at least 33% within organizations with over 100 employees), most nonprofits are too small to meet the threshold for tracking. This lack of timely, transparent information makes it difficult to support nonprofits and communities.
At Candid, we’re exploring ways to leverage technology to get a better picture of the current state of the sector. This involves two AI-powered processes. First, a machine learning model reviews incoming news articles and scores each one on how likely it is to describe a nonprofit closure or round of layoffs. Articles that clear a 50% confidence threshold are stored for monitoring, and those with a confidence rating of 80% or higher are passed to a large language model (specifically, Claude by Anthropic) to pull out key details: which organization is affected, where it’s based, when it happened, and what took place, whether that’s a closure, layoff, or both. Those details become structured data points that can help our data team better identify terminated nonprofits in the future.
What we’re seeing so far
We began exploring this methodology in March 2026, and while the tool is still very much in its experimental phase, we’re already seeing some patterns. The system reviews about 100 social-sector news articles a day. In its first three months, the tool surfaced close to 65 distinct potential closure or layoff events at U.S. nonprofits.
Some examples of nonprofit distress flagged by the tool includes the closure of Senior Resource Connection, an Ohio nonprofit that has been delivering Meals on Wheels and other senior services across multiple counties for over 70 years, and of North Napa Shelter, a homeless-shelter in California.
More holistically, a few patterns stand out in the data captured to date. Closures outnumber layoffs roughly seven to one, and non-federal funding cuts is the most-cited reason identified over the past three months. Specifically, 43 articles identified non-federal funding cuts as the cause of distress, and 40 articles mentioned that this led to a closure, while three articles noted that this led to layoffs (see chart below). The distress signals are also geographically dispersed across 28 states; no single state dominates.
Caveats and next steps
These are early findings, and it’s worth being clear about what they can and can’t tell us. For example, a flagged article is best interpreted as a “signal” of distress, not a confirmed event, as any AI model is imperfect. It is also important to note that we believe this methodology undercounts the actual number of nonprofit closures and layoffs, as smaller nonprofits are likely closing quietly, without making headlines. Therefore, data points indicate early warning signs of sector strain, not a complete count of what’s happening across it.
This project also highlights a critical information gap in the sector: Comprehensive data about nonprofit closures is needed to better support the social sector, especially in times of distress. But until we have that data, we’ll continue to experiment with new ways to gather data, measure accuracy, and scale the information.
Our hope is that this effort will sharpen the sector’s collective ability to see distress sooner and respond better. Early distress signals could help organizations plan ahead and help funders, boards, and policy makers better understand where emergency support is needed most. We also hope this experiment encourages both nonprofits and journalists to elevate news around nonprofit closures and other distress signals. After all, we can’t track what we can’t see.
Photo credit: skynesher/Getty Images
