RPA Bots in Federal Audits: Savings Claims and Security Gaps

TIGTA and the GSA inspector general audited government RPA bots. Findings on unverified savings, cost tracking, and bot access, with dates and figures.

Guide: AI Tools and IndustryLast verified: October 20266 min read

Federal inspector general audits of robotic process automation (RPA) found weak spots in savings claims, cost tracking, and security controls for bots, not in the quality of their decisions. The General Services Administration (GSA) inspector general reported on November 30, 2023 that a claim of more than 240,000 work hours reclaimed each year was inaccurate and unreliable, and on August 6, 2024 that none of 16 reviewed system security plans addressed bot access as GSA policy required. The Treasury Inspector General for Tax Administration (TIGTA) found in 2020 that an IRS bot saved far fewer hours than projected.

Key takeaways

  • GSA's inspector general found no verification of hours saved and no tracking of bot development costs, which left the claimed 240,000 hours unsupported.
  • In the GSA security audit, 55 of 56 bot custodians kept access to decommissioned bots beyond the 14-day limit in GSA's access policy.
  • TIGTA estimated that the IRS contractor responsibility bot saved 2,740 hours in its first year, against an IRS projection of 11,250 hours a year.
  • On June 18, 2026, the GSA inspector general closed its savings recommendations, concluding that corrective actions were appropriate.
  • None of the audits tested machine learning components, so they say little about AI accuracy.

What the auditors meant by a bot

The GSA inspector general describes RPA as software that uses bots to simulate human actions, such as copying data, filling in forms, signing into applications, and sending emails. A later GSA review calls bots rules-based software. TIGTA, in a 2020 audit of the Internal Revenue Service (IRS), used a second term, intelligent automation, for software that imitates human learning actions in tasks such as chatbots and natural language processing. The audited programs therefore mixed fixed-rule bots with a small amount of language technology, and the findings below mostly concern the fixed-rule part.

The IRS bot that was projected to save 11,250 hours

In May 2019 the IRS deployed the Contractor Responsibility Determination Robot, which gathers vendor data from federal websites to check whether a vendor is eligible for an award. The IRS projected savings of 11,250 hours a year, or about USD 1.35 million, at 2.5 hours per determination. TIGTA report 2020-20-060, issued on September 25, 2020, compared bot usage with 2,774 new contracts and estimated that the bot saved 2,740 hours, about USD 328,800, in its first year. That is roughly 24 percent of the projected annual hours, a ratio calculated for this article from the two figures.

TIGTA estimated another USD 485,400 in avoidable costs on 1,618 contracts where manual determinations were potentially made. The audit also noted that the program did not fully allocate direct and indirect costs to each automation project, so the true cost of a bot could not be determined. For example, USD 376,450 paid to a contractor for the bot came from another budget line. TIGTA did not independently validate the bot's accuracy.

The GSA savings claim

GSA's fiscal year 2020 financial report stated that its bots allowed the agency to reclaim more than 240,000 work hours every year. The inspector general's November 30, 2023 audit found that GSA was not verifying actual hours saved with the people who used the bots, and was not tracking bot costs, so it could not tell whether bots produced cost savings or a return on investment. The report recommended an evaluation process with auditable measures and cost tracking per bot. An implementation review dated June 18, 2026 found that the Office of the Chief Financial Officer had taken appropriate corrective actions and that no further action was necessary.

The GSA security audit

The inspector general's August 6, 2024 report covered 119 active and 24 decommissioned bots as of December 2022. It found that three monitoring requirements in GSA's own policy were not met: baseline monitoring of what data a bot touches, weekly log reviews by bot custodians, and annual bot reviews. Custodians had never been given access to bot logs. GSA then removed those requirements in a policy revision published in February 2023.

Of 16 system security plans reviewed, none was updated as the policy required, seven did not mention bots at all, and 10 failed to list the non-person accounts bots use. For decommissioned bots, 55 of 56 bot custodians kept access past the 14-day limit, and two custodians kept access to a bot that handled personally identifiable information for over four months. None of the 13 bot developers had access removed. GSA's finance and IT chiefs agreed with the seven recommendations but did not entirely agree with the finding. The status of those recommendations was not verified for this article.

Three audits side by side

AuditDateSubjectMain finding
TIGTA 2020-20-060September 25, 2020IRS contractor bot and chatbotSavings below projection, costs not fully allocated, weak governance
GSA OIG A210057November 30, 2023GSA bot program savingsClaimed 240,000 hours unsupported
GSA OIG A230020August 6, 2024GSA bot program securityMonitoring, security plans, and access removal fell short

What this means when AI is added to a bot

Vendors and agencies increasingly attach machine learning to bots, for example to read documents or classify messages. The audits did not test that layer. A reasonable reading is that the questions they raised carry over, because a bot with a learned component still runs under a non-person account, still touches records, and still needs a measured baseline for claimed savings. The checklist below summarizes those questions in general terms and is not a legal or audit standard.

  • Are hours saved measured with the people who use the bot, or only estimated?
  • Are development, license, and support costs tracked for each bot?
  • Does every system a bot touches list the bot and its account in its security plan?
  • Can bot custodians see the logs that policy tells them to review?
  • Is there a documented step that removes access when a bot is retired?

FAQ

Do these audits show that RPA does not work?

No. The TIGTA audit estimated real savings from the IRS bot, and GSA's inspector general closed its savings recommendations in 2026. The audits question how benefits were measured and how bots were controlled.

Are the audits about artificial intelligence?

Only indirectly. The bots audited were mostly rule-based, and TIGTA's chatbot was a minor part of its review. No audit tested the accuracy of a machine learning model.

Does the GSA security audit mean data was breached?

The report describes risk of exposure from excess access and gaps in monitoring. The material reviewed for this article does not report a confirmed breach.

Where can the full reports be read?

TIGTA and the GSA Office of Inspector General post their reports on their own sites, and the two linked above are free to download.

Last verified: October 2026

Sources

  • Process Automation Benefits Are Not Being Maximized, and Development Processes Need Improvement, Treasury Inspector General for Tax Administration, September 25, 2020
  • GSA Should Strengthen the Security of Its Robotic Process Automation Program, report A230020/B/T/F24004, GSA Office of Inspector General, August 6, 2024
  • GSA's Robotic Process Automation Program Lacks Evidence to Support Claimed Savings, report A210057/B/5/F24001, GSA Office of Inspector General, November 30, 2023 (as described in the 2024 report and the 2026 implementation review)
  • Implementation Review of Corrective Action Plan, assignment A260045, GSA Office of Inspector General, June 18, 2026

New to artificial intelligence?

Start with the basics, then follow the guides to tools, rules, and effects on society.

Start with AI Basics