Eleven Hours in Albany

The story begins not with a spy but with a phone call. In late July 2020, a person contacted the FBI’s Albany field office with a complaint. Asked later, formally, “How was source found?”, the case file answers in nine words: “The CHS called in a complaint.” [03]

CHS — confidential human source. The FBI met the caller on July 31 and opened him or her as a source on August 14. The Bureau’s own records describe the motivation in language that is almost poignant: “The source was motivated by a sense of indebtedness towards the United States after being granted asylum, and did not approve of the manner with which the PRC was governed.” [06] A Chinese asylee, grateful to America, hostile to Beijing, eager to help.

The source’s reporting stream, in its first weeks, looked like what an asylee plugged into a Chinese diaspora community might plausibly deliver. One early product was mundane and credible: a report about a Chinese talent-program recruitment invitation received at an American university. Albany wrote it up and released it as an intelligence report on September 22. Nobody objected. Nobody recalled it. It is still, as far as the record shows, in circulation.

The other item in the stream was different. On August 24, at the source’s second meeting with agents, he or she relayed something explosive: the Chinese government, in late August, had produced a large quantity of fraudulent U.S. driver’s licenses and secretly exported them to the United States, so that tens of thousands of Chinese students and immigrants sympathetic to Beijing could vote for Joe Biden despite being ineligible. The licenses, the source said, were built from the private data of millions of TikTok accounts — names, IDs, addresses — so they would carry real people’s information and be “difficult to detect.”

Where did the source get this? Not from personal knowledge. The FBI’s records trace the chain: the source heard it from a sub-source — a specific, named individual — who claimed to have heard it from “unidentified PRC government officials.”

When agents later pressed for details about this sub-source, the picture that emerged should give any intelligence officer pause. He was a retiree living in China, elderly, “cautious,” with a grown daughter. He communicated with the source over Telegram. He was, the source said, not a member of the Chinese Communist Party — in fact, he “came across as a supporter of the pre-1949 Chinese government, which coincided with Chang Kai-Shek’s regime.” A Nationalist sympathizer. A man whose politics placed him, by definition, outside the confidence of the ruling party — offered as a conduit to secret deliberations of that party’s government about a covert operation against the American presidency.

And the evidentiary trail to him had already vanished. The source “no longer had the last Telegram message exchange” with the sub-source “and believes that [he] may have deleted it.” The FBI added its own clinical footnote: “FBI-Albany confirmed that a security feature of the Telegram messaging application is the ability for any party to delete joint message exchanges.” They did not use email. There was nothing to examine.

The Bureau’s formal source description, printed in the report itself, compressed all this into a single garbled sentence: “A collaborative source with indirect access, none of whose reporting has been corroborated for less than one year.” Under “CORROBORATION,” the report’s administrative record says simply: “None.” [06]

One more number matters, because it later did a great deal of political work. When headquarters asked, “Scale of 1-10, how confident is the source in the reliability of the subsource’s information?”, the answer recorded was “9-10 range. ‘very very confident.’” In 2025, when these records reached Congress, that figure circulated as if the FBI had rated its source nine out of ten. It had not. The nine-out-of-ten was the source grading his own sub-source — the rumor’s carrier vouching for the rumor’s origin. The FBI’s actual assessment of the source, from the same document, was far more modest: “Just met the source, but was able to partially corroborate some of the information s/he provided. Specifically that someone the source identified to agents does in fact live in [redacted] and have contacts overseas.” The only thing the Bureau ever verified was that the man in China existed.

Set aside, for a moment, who said it, and consider what was said — and when.

Four weeks before the source relayed the license story, on July 27, 2020, U.S. Customs and Border Protection’s Chicago office published a press release announcing that officers at O’Hare’s international mail facility had seized 19,888 counterfeit U.S. driver’s licenses in the first half of the year, the majority shipped from China and Hong Kong. [09] The story ran widely through August. On August 6, President Trump signed an executive order against TikTok, and the app’s harvesting of American user data became the summer’s running technology-security panic. And the loudest election-year argument in the country, all summer, was about mail-in voting and fraud.

Chinese counterfeit licenses. TikTok data theft. Mail-ballot fraud. The Albany report is those three headlines, fused.

This is not an outsider’s cynical read. It was the FBI’s own, in real time. Three days after the report went out, the Albany analyst who drafted it wrote to colleagues: “the three main things the CHS stated (1. China’s trying to influence the election in favor of Biden 2. The threat of TikTok to US National Security and 3. China importing US driver’s licenses) were all documented in some fashion on open sources. Given the lack of specifics we received in the initial reporting, my first opinion was that the CHS wants to help and is probably supplementing his reporting via open sources.” [03] The first analyst to touch the claim suspected, from the start, that it was headlines wearing a trench coat.

Headquarters had the same instinct. Within hours of dissemination — before any recall discussion — two analysts at FBI headquarters’ counterintelligence division called Albany with their initial read: “they haven’t verified the alleged activity is occurring through other collection platforms, and believe it could be part of the PRC’s disinformation campaign toward the election.” Days later, headquarters asked Albany to put the question to the source directly: “is the source or sub-source aware of open source reporting concerning seizures of thousands of driver’s licenses at Chicago O’Hare? Further open sources began to suggest these drivers licenses were being used to vote fraudulently.” The handler asked. The source said no — the reporting came from “an individual in China,” and the sub-source “claimed that he did not obtain this information from any newspaper.” The denial is in the record. So is everything it is denying.

There was also a competing conspiracy theory in the same stream, and its provenance is instructive. Someone in the source’s reporting had claimed the Chinese government maintained underground bases in Republican states to disperse coronavirus. When a reviewing official asked whether that came from the same sub-source — “Is this subsource the same one who said that the Chinese Government had underground bases in Republican states to disperse Coronavirus?” — the answer dissolved the premise: “The coronavirus information was collected from twitter and has nothing to do with the election reporting … The source collected and documented everything he thought the agent may be interested in.” The FBI’s own file describes a source who fed agents Twitter content as intelligence. The eager asylee, helping.

Then there is the scheme itself, which fails on mechanics before it fails on sourcing. The plot requires physical counterfeit driver’s licenses to generate mail-in votes. But a mail ballot is not obtained by showing a plastic card. It goes to a registered voter, at a registered address, and registration is verified — under the Help America Vote Act — by matching a driver’s license number or Social Security digits against government records, then guarded downstream by signature verification. [16] A counterfeit card carrying a real person’s “true ID number,” as the report specified, collides with the real person’s existing registration. The one component of the scheme that is expensive, interdictable, and physically real — plastic cards shipped through a mail facility CBP was already searching — is the component least useful for mail-ballot fraud. It is, however, precisely the product required by the actual, well-documented market the CBP seizures served: American teenagers buying fake IDs to get into bars. CBP’s own examination noted the seized licenses repeated the same photographs across different names — the signature of a fake-ID mill printing lookalike cards for college kids, and the opposite of what identity-matched voter fraud would require. When the FBI’s task force later ran the lead to ground with CBP, the answer came back that there were no accompanying name lists, and that “the overwhelming majority of these documents are procured by teens between the ages of 18 and 20 … to circumvent alcohol laws.” [11]

The report’s own text contained the tell. Beneath the source’s claim that China had harvested addresses from millions of TikTok accounts, the FBI inserted a comment: “A persons address information was not a valid field when creating a TikTok account. It was unspecified how China would attain US address data from the application.” [01] The Bureau disseminated a scheme while footnoting that its enabling mechanism did not exist.

The report that went out at 7:23 that morning was not quite the report its drafter wrote.

The analyst who turned the source’s debrief into an intelligence report understood the sourcing problem and tried to flag it. Recognizing, as she later wrote, that recipients would never see the FBI’s internal source-evaluation data, she “elected to include” a context statement in the report itself: the information “passed through a sub-source who claimed contact with unidentified PRC government officials, leaving the veracity of information questionable but is being reported due to the nature of the threat being discussed.” [07]

That sentence — the drafter’s alarm, embedded in the product — did not survive review. “However this statement was requested to be reworded prior to release,” the analyst wrote afterward, “to ‘The source obtained the information from an identified sub-source who claimed contact with unidentified PRC government officials.’” The doubt was edited out. The reworded version is what the intelligence community, and the White House Situation Room, received.

Who requested the rewording? The file’s process chronology shows exactly one editing step between draft and dissemination: on September 22, Albany’s collection reports officer — the CRO, the official who owns what the office publishes — “provide[d] IIR edits” to the analyst. [02] The Bureau’s legal reviewers never touched the substance; their review, completed September 24 with “no objections/comments,” was scoped to a single question — whether the report could lawfully name a U.S. person. (It named Joe Biden.) Headquarters had no pre-release role at all; no rule required one. The rewording sits, on the evidence, with the same local official who would push the report out three days later — the office’s most pro-dissemination hand, softening the product’s self-doubt on its way to the door.

In fairness, the FBI’s rules invited the dissemination. Raw intelligence reports are governed by a deliberately low bar — “the dissemination threshold is lower for threat reporting,” the manual says, and it offers, as a model, exactly the kind of language the drafter tried to use: “The veracity of this source’s information is seriously doubted but is being reported here because of the nature of the threat discussed.” The system is designed to push uncorroborated threat reporting outward, caveats attached, and let the wider community evaluate it. Albany followed that design — minus the caveat. When headquarters called that first morning, Albany’s supervisory analyst summarized the office’s position with what reads, in hindsight, like tragic pride: “the IIR was coordinated and disseminated in textbook fashion … Something in the IIR process is definitely going well if we receive a call within a few hours from HQ about one of our IIRs. Nicely done!”

What happened next unfolded in one working day, and the file preserves it almost minute by minute. [02][04]

The report reached the FBI’s China-focused Foreign Influence Task Force at 10:19 a.m. By 11:59, analysts at headquarters were calling Albany: the report was “getting highlighted” inside the counterintelligence division. At 12:34 p.m. it reached Nikki Floris, the deputy assistant director for counterintelligence.

At 4:18 p.m., Floris emailed colleagues, including Tonya Ugoretz, her counterpart in the cyber division: “Are you guys tracking this IIR? Given the topic, this is obviously going to get quite a bit of attention. Thoughts on … adding in some additional analysts comments on this to put this reporting into context?” Ugoretz’s reply, at 4:24, was six words long: “Thanks. This is fun.”

At 4:35, a section chief in the intelligence directorate looped in the unit that manages raw reporting: “In addition to the IIR we discussed this morning, another election related IIR went out today that is creating a stir.” In addition to the IIR we discussed this morning — hold that phrase; it matters later. At 4:52, Bradley Benavides, section chief of the Foreign Influence Task Force, weighed in: “the NY and Albany FOs have pushed two IIRs in the past two days, neither of which should have gone out in my opinion,” adding that he had asked his task force to draft guidance requiring field offices to coordinate election-related intelligence with headquarters “prior to its dissemination.”

At 4:53, a supervisor recommended action, and his framing deserves precision: “[I]t’s first contact with a source. They were trying to get in front of the source a second time. It never should have gone out and we recommend an admin recall.” An administrative recall — the procedural category for technical and process defects. What ultimately issued, ninety minutes later, was a substantive recall — the category the FBI’s manual reserves for “significant concerns” about “accuracy of information or integrity of the source.” The paperwork escalated the recall’s meaning even as its stated rationale stayed modest; the recall notice itself gave the reason as the need “to re-interview the source,” and instructed recipients to “destroy all copies of the original report.”

Floris, at 4:55: “Awesome, thanks. So is AL going to recall this? Just FYI, I’ve hit up the DI about ALL election related IIRs having to come through HQ for review. More to follow on that!”

Albany resisted. At 5:15 p.m., the field office’s supervisory intelligence analyst pushed back: “We have no reason to recall at this point,” noting the handler had been trying all afternoon to re-contact the source. Two minutes later, the intermediary relayed the reality of the situation: “right after I got off the phone with you, I was contacted by Cyber DAD and CD DAD. They are directly requesting it be recalled until he is re-interviewed.” At 5:20: “I talked to the SIA … They aren’t looking to recall it, they believe it meets DRAIN” — the dissemination standard. At 5:21: “[t]his is moving fast. He just notified me they will recall it. Boom mic drop.”

Floris, 5:22: “AAAAAAH SNAPS!!!!! Well done … THANK YOU for handling!!!”

Benavides, 5:28, wanted the record clear: “Fact or suggestion for them to recall?”

Floris, 5:32: “Fact. Boom.”

At 6:08 p.m., the recall issued. Formally, recalls belong to field offices. This one belonged to headquarters — “Fact,” not suggestion — over a field office that had said, in writing, seventeen minutes earlier, that it saw no reason to recall.

The tone of the principals, preserved in their own Skype logs, needs no gloss. Floris to Ugoretz, that day: “another fucking IIR went out re election security.” “oh crap.” “how’s the sourcing on the new IIR?” “i don’t even know what to say anymore.” [04]

Three days later, on Monday morning, September 28, comes an exchange that, as of publication, has not been reported in full — a Skype conversation between Floris and a counterpart at another agency who supported the President’s Daily Brief process. [05] The counterpart asks: “do you have a serial for that IIR you mentiond to me last week re: China”. Floris: “it’s recalled.” The counterpart explains the interest: “talkign to PDB briefer / that’s why i’m asking.” The President’s briefers were chasing the recalled report. Floris supplies the serial. Then the counterpart adds: “she is also asking if there was a reevision on the russia one.”

The russia one. This is Benavides’s second report — the other of the “two IIRs in the past two days,” the one “we discussed this morning” of September 25, the New York product. It was Russia-related. And it had been handled differently: not recalled, but slated for “revision to the source statement” — a correction that, Floris discovers in real time, had never issued. “there should have been, yes,” she types. “but i haven’t seen it … let me check sentinel … doesn’t look like they’ve pushed out a revision yet / for fucks sake.” The counterpart’s reply lands like a thunderclap in a file otherwise full of process: “The Director has the original in his book this AM.”

The same Monday exchange contains the line that, ripped from context, became a 2026 headline. Floris types that she is pushing out a rule that day requiring all election-related reports to route through headquarters before dissemination — the reform Benavides had proposed — and then, wry and exhausted: “i’m basically running a shadow government across the FBI at this point.” The “shadow government” was a coordination memo. [15]

The recall’s stated purpose was to re-interview the source. The FBI did.

It took days to find him. The handler tried “two to three times” without a call back, resorted to “an anonymous email account,” and prepared to visit the source’s home “without attracting undue attention.” When contact came, the source stood by the story, named the sub-source, supplied the biographical details — the retiree, the Telegram, the deleted thread, the Nationalist sympathies — and professed himself “very very confident” in the man’s reliability. The information itself did not change. As Albany’s analyst put it to a colleague weeks later: “We did that, and learned more information regarding our sources sub-source, but for purposes of the information you are using, no information has changed.”

What happened next is the part of the story that the recall’s defenders have to sit with, because the file records it in the participants’ own words.

On September 29, a reviewing official at Albany framed the path forward this way: “Based on the conflict of this information to that of the wider USIC, the fact that it contradicts Director Wray’s testimony to Congress, and the importance to the current administration, I think we should be ready to back up the IIR with a minimum of the following information and treat it like it will be going in front of the President. We’re skeptical of this reporting based on the portions regarding COVID that veer into conspiracy theories.” The next day, the reports officer restated the requirements for reissue and closed with a sentence that would echo for years: “Again, the reporting will contradict Director Wray’s testimony.”

On October 1, the drafting analyst — the one whose veracity caveat had been stripped — wrote the most important email in the file. [07] “I do not agree with the reasoning for not putting this out,” she began. “Most concerning to me, is stating the reporting would contradict Director Wray’s testimony. I found this troubling because it implied to me that one of the reasons we aren’t putting this out is for a political reason, which goes directly against our organizations mission to remain apolitical and simply state what we know. Likewise, at the field operational level, I do not feel it is our job to assess whether or not our intelligence aligns with the Director, rather we provide intelligence for people way above our paygrade (like the director) to make the call of what the FBI is seeing as a whole.” She argued the field was not “in a sole position to determine a reporting’s validity” — that withholding it “takes away the crucial opportunity from the rest of the USIC to potentially corroborate or discredit our intelligence” — and proposed submitting a redraft to headquarters: “if they decide that they want to cancel it, let that be their burden.”

The official she was answering clarified, on October 2, that the Wray remark “was not to suggest we refrain from disseminating the IIR, rather to ensure everyone is aware of the attention this is likely to receive.” Perhaps. But the sentence had been written, twice, and an analyst on the receiving end had read it the way anyone would.

Albany tried the analyst’s path. On October 6, an acting reports officer — the regular CRO was out on COVID leave — sent headquarters a redrafted report. [08] It restored what the September version had lost, and more: “The source obtained the information from an identified sub-source, who claimed they obtained the information from unidentified PRC government officials. The access and placement of the ultimate sub-source of the information is unknown, but it is being reported due to the nature of the threat being discussed.” The drafter’s caveat, resurrected and strengthened. “If this addresses the concerns of DAD Floris and DAD Ugoretz,” the acting officer wrote, “then I will proceed to obtain OGC and LEGAT Beijing approval prior to dissemination; otherwise, please advise.”

Headquarters declined. On October 8, the task force’s unit chief closed the door: “we have not approved a reissue, specifically because of our concerns that the reporting is not authoritative.” Note what this proves in passing: the caveat-stripping of September was not why headquarters balked — offered the fully caveated version, they refused it too. The objection was to the chain itself, the retiree and the Telegram and the unidentified officials. On October 15, the analyst deleted the draft, citing “Internal Coordination.”

Inside the office, the private assessments were more candid than the official traffic. The acting reports officer, on Skype: “essentially i’m getting pressure to reissue it … I probably wouldn’t have put it out to begin with.” And: “I personally wouldn’t have put it out, but my CRO did and now no one wants to show up in an IG report … everyone is afraid of being viewed as partisan.” A task-force analyst pushed back on the partisanship frame from the other side: “its nothing to do with partisanship … its legit the information and we’d be trying to drill down regardless,” adding the mirror-image worry: “it could also come across as partisan putting the report out with such a strong title and no sourcing to back it up.” Both fears were rational. Both are in the file.

Meanwhile — and this detail undoes the tidiest version of the suppression story — the recalled report’s information kept moving through channels. On October 7, an analyst in the FBI’s Chicago division, working the CBP lead, asked Albany’s permission to use the report’s substance as a bullet in an election intelligence note, sourced to “a human source with indirect access and whose reliability is unknown.” Albany’s analyst agreed: “I have no issues with you using the bullet and think you have captured the information appropriately.” The claim was not memory-holed. It was circulating with its warning label — which is roughly what the drafting analyst had wanted all along.

And then, nothing. No reissue, and no investigation either. The task force ran the Chicago-CBP lead — teenagers, bars, no name lists — and stopped. When the FBI finally answered Senator Chuck Grassley in June 2025, it “found no information to indicate that FITF-China aggressively investigated the reported information, despite corroborating intergovernmental reporting and logical investigative leads.” [14] The Bureau had chosen the one combination with no defense: if the claim was potentially grave, it merited quiet investigation after the recall; if it was not credible, the recall’s paper trail should never have leaned on the director’s testimony. It did neither and both.

The recall might have stayed a footnote — one of 154 FBI intelligence-report recalls in 2020, a year in which Albany alone issued 86 reports and recalled exactly this one — had a whistleblower not carried it to Grassley, and had the political weather not changed. In 2025, the FBI under new leadership declassified the report and its correspondence for the senator. An internal Inspection Division team spent two weeks in July 2025 reconstructing the episode, reviewing 58,220 items from 103 accounts. Floris, by then departed, was “unable to contact”; Ugoretz sat for an interview. [02] In July 2026, the White House folded the whole package into its election-integrity declassification, where the recalled report stands as the collection’s only allegation of actual fraud — and the 16 July presidential address treated its suppression as proof of the crime. [12][13]

Here is what the record actually supports, after every document is read.

The claim is almost certainly false — not merely unproven. Its chain ends at an unreachable retiree whose politics disqualify the claimed access and whose one documented communication thread was deleted. Its content matches, element for element, the public headlines of the exact month of its collection, a derivation the FBI’s own analyst suspected in real time. Its mechanism was impossible as described, footnoted as such in the report itself, and contradicted by the physical evidence of the license seizures it almost certainly grew from. In five years, under a motivated Republican Senate inquiry and then an administration that seized county ballots and subpoenaed audit records in pursuit of exactly this theory, no corroboration of any element has surfaced. Georgia’s Republican-run citizenship audits, hunting the same genre of phantom, found twenty noncitizens registered among 8.2 million voters. [16] Tens of thousands of fraudulent Chinese mail ballots would have left a crater. There is no crater.

And the FBI’s handling was genuinely tainted — not fatally, but documentably. A drafting analyst’s honesty was edited out of the product by the office that shipped it. Headquarters commandeered a recall that policy assigns to the field, over the field’s written objection, inside eleven hours, with high-fives. The director’s day-old testimony appears in the reissue discussion twice, in writing, and a career analyst told her chain, contemporaneously, that it read as political. The substantive-recall label formally asserted concerns about source integrity while the stated rationale was a re-interview. A Russia-related companion report with its own source-statement problem stayed in circulation — uncorrected, in the director’s briefing book — while the China report died. And afterward the Bureau investigated almost nothing, leaving the allegation permanently unresolved, which is the condition in which allegations become legends.

These two convictions do not offset; they compound. Because the FBI neither reissued the caveated version nor investigated the claim to a documented conclusion, it left a vacuum, and in 2026 the vacuum was filled by the least accurate possible version of the story — the one in which the recall itself is the evidence. The one thing the file rules out most decisively — a genuine Chinese operation to manufacture tens of thousands of Biden ballots — is the one thing the report’s resurrection now asserts. Meanwhile the things the file actually establishes — a source supplementing his reporting from Twitter and the news; an editing culture that strips doubt from raw reporting; a headquarters that moves in hours when a report threatens the director’s credibility and not at all when it merely threatens the truth — went unmentioned in the primetime address.

There is a structural lesson here that outlasts the politics, and it is the reason this file deserves its place in the intelligence-community curriculum rather than the campaign reel. Finished intelligence — the assessments that reach presidents — passes through coordination, review, and dissent channels that, whatever their flaws, leave records and resist capture; when the 2020 election assessments were fought over, the fight itself was documented, the minority view published, the ombudsman engaged. Raw reporting has no such immune system. Everything that went wrong in Albany — the stripped caveat, the eleven-hour recall, the never-issued revision riding in the director’s book, the investigation that never happened — occurred in the space before finished intelligence, where the checks are thinnest and the file shows nobody was watching until it was too late, twice.

Nearly everything still contested about this story is answerable, cheaply, by people who already hold the documents.

The September 22 email in which Albany’s reports officer requested the caveat’s rewording exists in the FBI’s systems; the production to Grassley began three days too late to capture it, a scoping artifact that one further production would cure. The “russia one” has a serial, a subject line, an originating office, and a revision that either issued or did not; the FBI knows all four. The sub-source’s claimed pedigree sits under two redaction bars. And the question of whether any element of the license claim ever found corroboration is a records check the Bureau has apparently already run — the answer, per its own 2025 letter, being no. [14]

An afternoon of declassification review would convert this story’s remaining inferences into facts. The government that released fifty-eight documents to prove an election was stolen could release four more and prove, instead, how a rumor becomes a file, how a file becomes a recall, and how a recall — mishandled twice, by two administrations with opposite motives — becomes whatever the loudest reader needs it to be.