After raiding the election offices of Georgia’s most populous county last month, the FBI has disclosed information indicating where its investigation is heading.

Federal laws may have been broken during the 2020 election, according to the affidavit supporting the court-approved raid. Yet the breadth of the materials seized shows that the FBI may be able to check the integrity of the ballots more broadly, uncovering further issues or putting speculation to bed.

President Donald Trump’s campaign challenged the Georgia election most vigorously, as he lost the state to President Joe Biden by fewer than 12,000 votes, according to the official tally. The legal challenges failed. Instead, Trump was indicted based on the rationale that his efforts to challenge the election results were allegedly executed with corrupt intent. The case was dismissed after he became president again in 2025.

The renewed investigation now targeting Fulton County, Georgia, which covers the broader Atlanta area, uses a rationale analogous to that of the case against Trump. The affidavit states that if known irregularities in the election were intentional, such acts would be criminal.

On Jan. 28, agents seized some 700 boxes of election records, including physical ballots from the 2020 election. County officials have since filed a lawsuit seeking to have the materials returned.

The issues detailed in the affidavit were largely discovered years ago by concerned citizens using data obtained through Freedom of Information Act requests or litigation. Georgia Secretary of State Brad Raffensperger, who was responsible for overseeing the election and is running for governor of the state, has dismissed the issues as administrative and human errors too small to affect the election’s result.

However, the FBI has a different perspective.

“If these deficiencies were the result of intentional action, it would be a violation of federal law regardless of whether the failure to retain records or the deprivation of a fair tabulation of a vote was outcome determinative for any particular election or race,” reads the affidavit signed by FBI Special Agent Hugh Evans.

Raffensperger has repeatedly stressed that the 2020 votes were counted three times, including recounts done by hand and by machine.

However, many of the deficiencies outlined in the affidavit happened during these recounts.

The Original Count

Vote counting in Georgia starts by law on Election Day. Fulton County had more than half a million ballots to tabulate—almost 90 percent cast early or by mail. The result was announced several days later: Biden won the county by a 26-point margin.

One issue with the results was a lack of receipts. Each tabulator machine is supposed to be “closed” when polls close and tabulator tape is to be printed out to show how many ballots and votes for each candidate were counted. Then, the tape should be signed by the poll manager and two witnesses.

Yet tabulator tapes for more than 300,000 votes were not signed, and some were missing altogether, Evans wrote, referring to an analysis by Clay Parikh, a voting machine security expert.

Raffensperger said that that was merely administrative oversight, as the vote tallies are not recorded on the tape alone. They are also preserved on memory cards in the machines.

But Parikh’s analysis went deeper.

“Parikh identified one tabulator that was used to close out 15 tabulator machines from 12 different locations,“ Evans wrote. ”In addition, the poll closing time and report printed times on several closing tabulator tapes were close enough in time that Parikh believed someone had to have manipulated the times on the reports.

“Parikh believed this showed that the memory cards were removed from the original tabulator and put in another tabulator to print out the closing tabulator tapes.”

The tabulators also have “protective counters” that track how many ballots have been scanned on them over their lifetime.

“[Parikh found that] the protective counters on at least five tabulator tapes from the same unit were identical,” Evans wrote. “Some of the reported ballots scanned exceeded the protective counter number.”

“This indicated to Parikh that no ballots were ever scanned on these machines and that the numbers generated from those ballots were done so by placing an unencrypted memory card into the unit to generate the closing tape,” Evans wrote.

“This would have allowed an opportunity for the tabulation to be tampered with.”

The tabulators are supposed to scan each ballot, creating a digital record. But the majority of the images from the original in-person voting count have not been preserved by the county, Evans said. At the time, the county was not legally required to preserve them, but it is not clear why they were discarded to begin with.

“This is another impediment to ruling out non-criminal explanations for the activities during the election,” the affidavit states.

Hand Recount

On Nov. 11, 2020, Raffensperger announced a risk-limiting audit. Because the race was so close, state law required all the ballots to be recounted by hand. The ballots were counted in batches and the final tally for each batch was supposed to be put into an electronic auditing system called “Arlo.”

Several people who participated in the audit said they witnessed suspicious occurrences, including a batch of 110 ballots that contained 107 featuring votes for exactly the same candidates. The bubbles on them were filled in exactly the same way and their paper felt different from that of other ballots, the participants said. The ballots were marked as absentee but lacked creases from being folded in a return envelope.

It is possible that such “pristine” ballots can be created by duplication, during which a damaged ballot is copied onto a new one. But those should be clearly marked as “duplicate,” and the original needs to be preserved, Evans said.

One of the witnesses, who had been a poll manager for 25 years, also said she remembered a batch of about 60 ballots marked as coming from a senior living center.

“[She] believed these ballots should have been folded as well but were not,” the affidavit states.

Yet another witness, one of the Fulton County commissioners, was a poll worker at the time. When helping test the voting machines before the election, she saw a pile of unsecured papers used to print testing ballots.

“She stated [that] she could have printed any ballot she wanted,” Evans wrote.

She also said she saw some people “printing random ballots” and managed to rip some up, according to Evans.

“She was not sure the reason they were printing ballots as all the test ballots had already been printed,” he wrote.

None of the witnesses in the affidavit were identified by name.

Evans also mentioned a complaint submitted to Georgia Gov. Brian Kemp by chemical engineer Joseph Rossi, alleging inconsistencies in the hand recount results for dozens of ballot batches. Kemp’s office independently verified the allegations, concluded that they were factual, and passed them on to the State Election Board for an investigation, which was eventually conducted by Raffensperger’s office.

Raffensperger dismissed those as human errors during data entry. But some of them raise the question of how such a specific error could have been made.

For example, one batch was reported as 200 votes for Biden and zero for any other candidate. But when Kemp’s office checked the ballot images for that batch, it showed 85 votes for Biden, 12 for Trump, and three for other candidates.

Another batch was reported as 150 votes for Biden and zero for other candidates. In fact, the batch contained 97 votes for Biden, eight for Trump, and one for a third-party candidate.

There were two more batches reported as each having 100 votes for Biden and zero for others. In fact, one had 87 votes for Biden and 10 for Trump; the other had 74 for Biden and 25 for Trump.

Machine Recount

On Nov. 21, 2020, the Trump campaign requested another recount. Based on state law, this one was conducted by rescanning ballots through the tabulation machines.

Again, concerned sleuths found problems. When they analyzed the ballot images, they found that more than 3,000 were duplicates. Upon investigation, Raffensperger’s office could not tell for sure whether those ballots were double counted. If they were indeed counted twice, it would have lowered Biden’s margin of victory. Because of that, Raffensperger’s investigators concluded that it was not “intentional misconduct” and did not investigate further, according to the affidavit.

One of the independent researchers who identified the issue argued that “the introduction of duplicate ballots was intended to make the recount numbers match more than to affect the outcome of the election,” the affidavit states.

“Such action would be an intentional tabulation of ballots in a false manner” and therefore criminal, Evans said.

The researcher also found that more than 17,000 ballot images from the recount were missing.

The county was not required by law at the time to preserve the images, but the researcher “concluded it was suspicious that the missing image files were deleted at approximately the same time the duplicate votes began appearing in the system,” the affidavit reads.

Under the Microscope

The 2020 Georgia election has been one of the most scrutinized, including through recounts, state investigations, independent investigations, performance reviews, and now a federal criminal probe.

None of the probing so far has uncovered definitive proof of fraud that would substantially affect the results, despite finding a plethora of problems.

Evans also mentioned a report by Seven Hills Strategies, an Atlanta-based consultancy tasked by the State Election Board with monitoring pre-election work in Fulton County.

The report concludes that the county’s handling of absentee ballots was “extremely sloppy and replete with chain of custody issues.”

Evans wrote that during the hand recount, there were “chain of custody issues, ballots left unattended, unsealed bags being used, and auditors not recording seal numbers on ballot bags.”

The Seven Hills report identifies no deliberate misconduct.