There has been a lot of coverage this year about the IRS losing staff. Between DOGE-related cuts, buyouts, and layoffs, the agency went from roughly 102,000 employees in early 2025 to about 74,000 by December. That’s a 27% reduction. The Large Business and International division handles most R&D credit examinations for mid-size and large companies, and it absorbed some of the heaviest cuts. LB&I lost a significant number of its most experienced revenue agents, the ones who had spent years working complex credit claims.
Clients keep asking us some version of the same question: does a smaller IRS mean less risk on the R&D credit?
No. And if you’re counting on the IRS being too stretched to scrutinize your claim closely, that’s not a strategy worth betting on.
Fewer Agents, More Selective Enforcement
When an agency loses a third of its workforce, it doesn’t abandon enforcement. It gets more deliberate about where it focuses its remaining capacity. For the IRS, R&D credits have always been a priority target. Not because the credit is common, but because the potential adjustments are large. A single examination of a mid-size company claiming a few million dollars in credits can produce significant revenue recovery. A leaner agency needs every dollar it can find, and it will go where the money is.
Bloomberg Law’s January 2026 coverage of the current IRS posture made this point directly: the R&D credit will remain a priority through the staffing reduction precisely because the financial stakes are high. Complexity combined with high potential adjustments keeps a credit on the examination radar year after year, regardless of how many agents the agency has.
There’s also a political dimension here. The IRS has faced public scrutiny over whether its enforcement capacity has weakened. Agency leadership has responded by emphasizing that enforcement will continue on high-value, high-complexity items. R&D credits fit that description. Walking away from them would send the wrong signal.
The LB&I Backlog
The staffing losses in LB&I have created a real bottleneck, but not in the way most companies hope. Audits aren’t disappearing. They take longer to assign and longer to resolve. Once an examination opens, it stays open longer. That means more time, more disruption, and more document management on your end. It also means more opportunity for scope to expand during the examination, since examiners have more time to look at additional issues while a case sits in the queue.
The remaining agents are also changing how they conduct examinations. Practitioners who work regularly with exam teams report less tolerance for inferential reasoning. Agents now require more direct evidence that activities qualify under the four-part test. They want documentation that connects specific work to specific individuals, specific expenditures, and the actual process of experimentation, not just project summaries. The agents who remain in LB&I have seen enough examinations to spot documentation assembled for the audit rather than created during the work. That bar keeps rising.
Data-Driven Enforcement
A smaller workforce doesn’t mean less scrutiny. It means the IRS relies more heavily on technology to do what people used to do. IRS leadership has explicitly called data-driven enforcement one of the agency’s top strategic priorities going into 2027. Automated screening, analytics, and pattern recognition now handle more of the front-end work of identifying which returns to flag for examination. Fewer agents can generate the same number of examination referrals once the selection process runs on automation.
Form 6765’s new Section G was built with exactly that in mind. Once companies report R&D data in a standardized, project-level format, the IRS can run screens across thousands of returns and flag outliers without a human reviewing each one. Unusually high credit-to-wage ratios, business components that don’t fit industry norms, allocation percentages that look like estimates rather than measurements: all of that becomes visible at scale. Section G stays optional for tax year 2025 but becomes mandatory starting in 2026. We have a detailed breakdown of what Section G actually requires and how to build toward it.
Software Development Claims Are Getting Extra Attention
One area deserves specific mention: software development R&D claims in financial services are seeing significantly more examination activity. Banks, investment management firms, and fintech companies that claim credits on software development work face growing scrutiny. Those examinations tend to run intensive.
The tension sits between the IRS’s traditional model of research and how modern software development actually works. Agile sprints of two to four weeks scope each sprint based on the results of the last one, and that represents a genuine process of experimentation. The uncertainty is real. The alternatives under evaluation are real. But documenting that process in a way that satisfies an examiner thinking in terms of hypothesis-driven laboratory research takes intentional documentation infrastructure. If this describes your company, build that infrastructure during development, not when an examination notice arrives.
What to Do With This
The practical implication here: the risk environment for R&D credits hasn’t softened. The IRS has grown more selective about where it directs examination capacity, but more determined to produce results on the cases it selects. Automated screening, reduced tolerance for inferential documentation, and mandatory project-level disclosure starting in 2026 push the standard for a defensible claim higher than it has ever been.
A smaller IRS isn’t a safe harbor. The companies that get through R&D credit examinations cleanly aren’t the ones with the smallest claims, and they aren’t the ones who filed hoping no one would look. They’re the ones whose documentation stands on its own: records created during the work, business components described with specificity, and wage allocations tied to actual data.
Want to talk through where your R&D credit process stands? Reach out. We would love to help.