Run Correctly, Disbelieved Anyway: Eleven Ways the November 3 Election Can Go Wrong

If you read the news about the 2026 election long enough, you come away with the impression that something enormous is about to happen to it. Two executive orders, a Postal Service rule, a citizenship database, a hundred lawsuits, agents at polling places, a Supreme Court case about mail ballots. It reads like an election under siege.

Here is the strangest fact about all of it: as of the end of September, almost nothing has actually changed for voters. The executive orders are blocked. The Postal Service rule was denied by the Supreme Court nine days before the deadline to mail overseas ballots. The citizenship database is under challenge in Maryland. The Justice Department has lost every voter-roll case that has been decided, twenty-four of them in a row. The SAVE Act never passed. The Election Assistance Commission cannot adopt new standards because it has no commissioners left to vote.¹

Meanwhile, public confidence in the outcome has fallen from 77 percent after the 2024 election to 60 percent in January to 55 percent in September, and only 49 percent of Americans expect the midterms to be free and fair.²

That is the paradox worth sitting with. Everything designed to change the rules is losing. Everything designed to change the story is winning. This election is not at risk of being stolen in the sense that phrase usually means. It is at risk of being run correctly and disbelieved anyway.

Understanding that split is the whole exercise. So let me map the risks in order of how likely they are to matter, and be precise about which ones are documented, which are litigated, and which are only predicted.

Risk 1: The results are not believed

This is the highest-probability risk in the field, and the hardest to see because it happens after the votes are counted.

The numbers are stark. UC San Diego’s Center for Transparent and Trusted Elections has tracked confidence in the national vote count through the year: 77 percent after 2024, 60 percent in January 2026, 55 percent by September 15. Only 49 percent expect the midterms to be free and fair. And 89 percent of respondents hold their confidence level, whatever it is, with “at least moderate certainty.” These are entrenched views. They will not move on evidence.²

You can watch the mechanism work in miniature. In Wisconsin’s August primary, Milwaukee’s central count ran smoothly for about fifteen hours, and then a tabulator export error delayed results roughly ninety minutes past midnight. Within minutes, a lawyer and a podcast host were posting that the election had been cheated. There was no evidence of tampering and no sign the results were affected. It did not matter. The delay was the evidence.³

That mechanism has a name now. The Election Truth Alliance, a group that formed on Reddit after 2024, uses a University of Michigan political scientist’s statistical model to claim that fraud in the 2024 election exceeded Donald Trump’s margin in Pennsylvania, and that North Carolina saw more than 200,000 fraudulent votes. Ten election and statistics experts told Votebeat the model produces false positives and cannot detect fraud in American elections; the researcher himself calls it “an imperfect measure” and says the group has mischaracterized his work.⁴

Two things make this narrative worth watching closely. First, it targets 2024, which Trump won. This is not the 2020 grievance wearing a new coat. The fraud-claiming instinct has moved off the result it was built for and onto the next one, which means it is not a factional position so much as a habit. Second, and more important, it does not need to be believed by a majority to do damage. It only needs to be believed by enough people in the right counties to delay a canvass, force a recount, or make a certification vote politically expensive.

The Brennan Center’s Sean Morales-Doyle put the strategic logic plainly: the president “has no more power to run our elections than the activists bringing these challenges. So the only way their efforts to undermine our elections will succeed is if voters lose faith in the whole enterprise and stay home this fall.”⁵

That is the target. Not the count. The belief that the count means anything.

Risk 2: The money lever

If there is one mechanism in this whole landscape that is actually working, and that almost nobody has blocked, it is this one, and it deserves more attention than it has received.

On July 10, the Department of Homeland Security, acting through FEMA, attached five conditions to the Homeland Security Grant Program. The program is worth $1,064,000,000 across 56 awards. FEMA will withhold 20 percent of each recipient’s award until the state submits proof of compliance and the department verifies it.⁶

The five conditions: submit a plan to move away from voting systems that count ballots by bar code or QR code in favor of hand-marked paper; conduct a manual audit of at least 5 percent of ballots cast after each federal election; reconcile the number of voters who participated against the number of ballots cast; within 120 days of accepting the award, run the state’s entire voter registration database through the SAVE citizenship-verification system; and verify the citizenship of everyone working at polling places or operating election systems.

Read that list carefully, because it is doing two different things at once. Manual audits after an election and ballot reconciliation are good practice; they are what serious election administrators already do or want to do. Hand-marked paper ballots are the position of many of the people who have spent years warning about voting-machine security. Those three conditions are uncontroversial.

The other two are the SAVE Act, implemented by administrative condition, four months after the Senate declined to pass it. The bill would have required documentary proof of citizenship to register and would have run voter rolls against federal databases. It died. The grant condition does it anyway, and it adds something the bill did not contain: citizenship screening for poll workers, which reaches “all individuals involved in the administration of elections.”

The legal theory is different from everything that has failed so far. The executive orders died because the president has no authority over elections under Article I, Section 4. Conditioning a voluntary grant program is not regulating elections. It is saying that if a state wants the money, it will do these things. States can refuse. Some are considering it. And the footnote in the DHS release, acknowledging the June 2026 court order that vacated the SAVE expansion and offering an alternative immigration records search, shows that the department is lawyering around the injunction rather than ignoring it.

The same lever appears elsewhere. DHS has threatened to withhold emergency preparedness and counterterrorism funding from states that refuse to implement White House election policy. Secretary Markwayne Mullin told four secretaries of state, in writing, that they should “confirm their intentions to collaborate with DHS” within two weeks.⁷

Why this matters more than the headline fights: it is the one channel that does not require winning in court. It requires only leverage over state budgets, and it moves quietly, on a grants calendar, away from the cameras.

Risk 3: Certification becomes a choke point

This is where the structural fight has actually relocated, and the good news is that it is the best-defended part of the system.

Certification is the step where election officials sign off that the count is complete and accurate. It has always been a formality. Since 2020 it has become a target, and the shape of the threat has changed: what election deniers learned after 2020 is that outright refusal fails, but delay is useful. A delayed certification amplifies conspiracy theories, creates the appearance that something is being hidden, and forces courts to spend time on procedural fights instead of substance. In a close race, a delay can outlast a news cycle.

The dates are knowable and worth knowing. In the seven battleground states, certification must be complete by: Arizona, November 23; Georgia, November 20; Michigan, November 23; Nevada, November 24; North Carolina, November 24, with a possible adjournment to December 4; Pennsylvania, November 23 at the county level, with no statutory deadline at all for the secretary of the commonwealth; Wisconsin, December 1.⁸

Two things in that list are worth flagging. Georgia’s deadline is the earliest, and its counties certify first, on November 9. And Pennsylvania, one of the largest and most closely watched states, has no deadline for the state’s own certification. Nationally, only Pennsylvania and Rhode Island lack one.⁹

The defenses are real and they have worked every time. State law uses mandatory language, and courts read it that way. In Michigan the constitution itself says certification is “the ministerial, clerical, nondiscretionary duty of a board of canvassers.” Georgia’s Supreme Court struck down a State Election Board rule that would have let certifiers delay for an undefined “inquiry.” A Fulton County court wrote that local officials “must certify and must do so by a time certain. There are no exceptions.” Virginia’s courts held that “the personal beliefs of members of a local board of elections cannot derail the electoral process for the entire Commonwealth.”¹⁰

Two states can substitute for a refusing county. In Michigan, if a county fails to certify, it must hand over its records and the state board completes the canvass, with all costs borne by the refusing county and the entire county board required to sit present throughout. North Carolina’s state board can seize missing abstracts from the clerks of court and remove county board members, which it did in 2023. New Mexico gives voters themselves a private right of action.

Writs of mandamus have worked in Arizona, New Mexico, Nevada, Virginia, and Pennsylvania. Arizona prosecuted two Cochise County supervisors who refused in 2022; one pleaded guilty to a misdemeanor.

So the honest assessment is that a certification refusal in 2026 will almost certainly fail. The risk is not that it succeeds. It is that it costs a week, feeds a narrative, and provides a hook for the bigger claim that the whole election was tainted.

One caveat worth carrying: Arizona, of all states, is currently loosening this framework. Its legislature passed a bill in 2026 that would rewrite the duty so supervisors merely “acknowledge without prejudice” rather than canvass. It passed the Senate and stalled on a tie vote in a House committee. A state court also struck the election-manual language calling the county canvass “non-discretionary,” holding the manual goes beyond the statute. Arizona’s reputation as the most robust mandamus state is less settled than it looks.

Risk 4: The winner is not seated

This is the last step in the sequence, and it is the only rules-based risk in this piece that no court has closed.

Count, then certify, then seat. Risks 1 through 3 cover the first two steps. The third happens on January 3, 2027, when the 120th Congress convenes, and it turns on the one part of Article I that the Supreme Court has read narrowly in one direction and never in the other.

The rule. Article I, Section 5: “Each House shall be the Judge of the Elections, Returns and Qualifications of its own Members.”¹¹ For most of American history that sentence was read as a grant of broad discretion. On the qualifications half, it is not.

Powell v. McCormack, 1969. The House refused to seat Adam Clayton Powell Jr. after his re-election, on findings that he had diverted House funds and filed false reports. The exclusion passed 307 to 116, well past the two-thirds that expulsion would have required. The Court struck it down anyway, because the Speaker had ruled the vote an exclusion rather than an expulsion, and the Court held the two are “not fungible proceedings.” The holding: a member duly elected who meets the three Article I qualifications of age, citizenship, and inhabitancy must be seated, and “the Constitution does not vest in the Congress a discretionary power to deny membership by a majority vote.”¹² U.S. Term Limits v. Thornton confirmed in 1995 that those qualifications are fixed and “may not be supplemented by Congress.”¹³

A bare refusal to seat a certified winner is therefore unconstitutional. That part is settled.

The part that is not settled. Powell narrowed one word: “Qualifications.” It said nothing about “Elections and Returns.” Which candidate actually received more lawful votes is a different question, and in Roudebush v. Hartke the Court called it “a nonjusticiable political question,” beyond the courts even before the Senate acted. Powell’s footnote 44 extends the qualifications limit to the Senate identically.¹⁴ The elections and returns prong it left alone.

The consequence is an asymmetry worth stating plainly. The door is closed on “we do not like you.” It is open on “we do not believe the count.” A refusal dressed as a dispute over which candidate won sits in the one half of Article I, Section 5 that no court has limited. That is a story claim wearing a rules costume, which is the fusion this piece keeps returning to.

Powell also reserved more than it decided. Footnote 41 declines to say whether Section 3 of the Fourteenth Amendment, the impeachment-disqualification clause, the incompatibility clause, or the Article VI oath are “qualifications” within the clause. Footnote 42 leaves open whether a court may review a pretextual finding that a member failed a qualification. And the Court granted only declaratory relief, leaving to the lower courts whether a seating can ever be judicially compelled. Those are open doors, not rulings, and none has been walked through.¹⁵

The mechanical surprise. The popular version of this scenario gets the order backwards. The Speaker is elected before anyone takes the oath. Unsworn members-elect vote for Speaker; that is settled practice, and it is why the 2023 Speaker fight ran fifteen ballots with nobody sworn until it ended. There is no version of refusing to swear enough members that prevents the House from organizing.¹⁶ The only chokepoint upstream of the Speaker vote is the Clerk’s roll of members-elect, which the statute conditions on credentials showing the person “was regularly elected.” Striking a certified winner there would contradict the provision’s own text, and it has never been tried.

And the arithmetic runs the other way. At 218 Republicans and 214 Democrats, with control resting on a net of three seats, refusing minority members would strengthen the majority, and refusing the majority’s own members would strengthen the minority. Four refusals produce a tie. Five flip floor control. The mechanically dangerous case is the legally foreclosed one. The legally arguable case is not party-correlated, because no majority has a self-interested reason to refuse its own winners.¹⁷

What exists in 2026: nothing. That is the finding, and it is why this risk is ranked here rather than higher. No member, no leader, and no Speaker candidate has said they would refuse to seat November’s winners. No bill touches seating; the major 2026 election bill, H.R. 7300, does not mention it.¹⁸ The rules package for the 120th Congress has not been adopted, and seating was not raised at the September Member Day hearing.¹⁹ The mechanism that does exist has been used as a filter rather than a weapon: the 119th Congress dismissed every election contest that reached it, all of them without objection in December 2025.²⁰

The scenario appears in 2026 only as forecast, from an election-law podcast, an advocacy organization, a legal news outlet, and one newspaper column. None of them quotes an officeholder proposing it. The Brennan Center’s September briefing, which catalogs election-subversion worries at length, does not raise seating at all.²¹²²

Two traps in the discourse. The first is Grijalva. In autumn 2025 Speaker Johnson did deny a certified winner her seat for fifty days, the longest such gap in American history, and it is tempting to offer as precedent. It is not one. Johnson did not dispute that Adelita Grijalva won. He declined to convene the House. That is a scheduling refusal, not an election contest, and conflating the two is the easiest mistake available here.²³ The second is the Senate, which is structurally different: a continuing body that does not re-adopt rules and has no organization-day crisis, with no statutory equivalent to the House’s contested-election act, which is House-only by its own definitions. The Senate last refused to seat anyone in 1929.²⁴

The honest weight. As law, a refusal to seat a certified, Article I-qualified winner is foreclosed. Every historical precedent for it, the House exclusions of 1868 through 1920 and the Senate’s 1929 refusal, predates Powell. Since then the House has adjudicated 107 contested elections and never excluded a qualified winner.²⁵ As a 2026 forecast, it is a named scenario with no documented adherents among the people who would have to execute it. That is a different evidentiary shape from every other risk in this piece. Most of them have a filing, a directive, or a vote behind them. This one has commentary.

Which is the reason it sits at number four and not higher, and also the reason it belongs in the list at all. The exclusion of this risk rests on the empty 2026 record, not on the strength of the constitutional argument. A single documented statement would move it up.

Risk 5: Federal agents at the polls

This is the scenario that dominates the speculation, and it is worth being precise about what has actually happened and what the law says.

Three documented incidents, all during primaries. In May, during early voting in Bexar County, Texas, nine ICE agents swarmed a car in the parking lot of a library that was serving as a polling place. The county sheriff went to the scene and told them to leave. In June, during the California primary, ICE agents were seen in the parking lot of a polling place in Simi Valley, and park rangers persuaded them to depart. Later that month, during New York’s primary, ICE agents entered a library in Syracuse that was being used as a polling site, looking to interview a poll worker about her social media posts in an investigation unrelated to the election. No voters were present, and the county elections commissioner rushed to the site.²⁶

In none of the three cases is there any indication the agents were targeting election sites. In all three, local officials moved quickly and the agents left. That is the reassuring part, and it is also the point: the effect was fear in those communities, and the incidents were handled by local officials improvising under pressure, not by a system designed for it.

Two more 2026 data points. On primary day, June 9, a sitting South Carolina state representative became the subject of a felony investigation for displaying a firearm during an altercation at a polling site in Colleton County; the charge carries up to five years.²⁷ And on September 10, Common Cause, LULAC, UnidosUS, and the City of Denver sued in federal court seeking a declaration that armed federal agents at polling locations are unlawful under Section 592. Denver’s county clerk has begun training staff on de-escalating encounters with federal agents and identifying plainclothes agents; the complaint notes the threat is already harming poll-worker recruitment.²⁸

What has not happened, and it matters: I found no documented 2026 case of armed private individuals or militia members at a polling place. Armed presence at drop boxes and polling places is a 2020 and 2022 phenomenon and a 2026 forecast, not a current observation. If you see that claim made, it is being asserted without evidence.

The law is unusually clear here. Federal statute has barred armed federal personnel from polling places since the Civil War era. The relevant provisions: 18 U.S.C. §592, which criminalizes deploying “troops or armed men at any place where a general or special election is held,” and whose original exception for keeping the peace was repealed more than a century ago, meaning the administration cannot justify deployment by claiming local police are failing. Section 593 bars military personnel from using “force, threat, intimidation, advice, or otherwise” to prevent voting. Section 594 bars anyone from intimidating a voter. Section 595 bars federal employees from using official authority to interfere in a federal election. The Posse Comitatus Act separately bars the regular armed forces from domestic law enforcement absent congressional authorization.²⁹

Two enforcement details are worth carrying because they are rarely mentioned. These are five-year statutes of limitations, so an official who complies with an unlawful order today can be prosecuted later. And state prosecutions are not extinguished by a presidential pardon or a Justice Department declination. California and Colorado create a presumption that someone carrying a weapon near protected election activity is intimidating, and that presumption does not exempt federal agents.

On the 1,000 monitors. The Justice Department’s Civil Rights Division announced 1,000 election monitors for the general election, against nine jurisdictions in 2022 and twenty-seven in 2024. The Brennan Center’s assessment of what monitors may actually do is worth quoting in full, because it deflates the number: congressional observers may watch and question officials but may not handle ballots or touch equipment, and state and local officials may remove them for intimidation or disruption. Federal observers under the Voting Rights Act require a court order, and only three jurisdictions currently qualify. DOJ staff monitors have no statutory authority for special access; in many states they cannot enter a polling place without state or local consent, they are subject to state observer law, and they can be removed for violating election law.³⁰

So 1,000 monitors is a large number attached to a small set of powers. The number matters for what it signals. The powers matter for what it can do.

And the counterweight. The chairman of the Joint Chiefs of Staff, Gen. Dan Caine, wrote to Sen. Elissa Slotkin on August 28 that the military has no plans to send troops to the polls in November, and that “election administration and security remain the fundamental responsibility of state and local officials.” The Associated Press and NBC News each obtained the letter independently, and the Pentagon said it had nothing to add beyond its contents.³¹ The Supreme Court also recently blocked the administration from federalizing National Guard troops during the immigration surge in Chicago. And it remains within each governor’s discretion whether to call up their state’s Guard at all.

Risk 6: The data dragnet

The most sustained federal campaign of the cycle has been the attempt to assemble voter data at national scale. It has lost in court more completely than anything else here, and it has still partially succeeded through voluntary compliance.

The record: 0 for 24. The Justice Department demanded full statewide voter rolls, including driver’s license numbers and partial Social Security numbers, from 48 states and the District of Columbia, and sued 30 states and D.C. when they refused. It has lost every case that has been decided. The Sixth Circuit affirmed against it. Judges have been unusually direct about motive: an Oregon court found the department’s inconsistent arguments and contradictory public statements evidenced “ulterior motives,” and that its word could no longer be presumed regular. A Vermont court noted that the department admitted in a hearing it had no evidence the state was out of compliance with anything, because it was “still in the investigatory phase.”³²

The legal failure has a technical shape worth understanding, because it explains why the loss is so total. The department’s authority comes from Title III of the Civil Rights Act of 1960, which lets the attorney general demand election records but requires a statement of “the basis and the purpose.” Courts have held either that the department failed to state a basis, or that voter rolls, being built from information voters supply, are not records that “come into possession” of election officials in the sense the statute means. Two independent grounds, and the department loses on both.

But 16 states handed over full unredacted lists. Alabama, Alaska, Arkansas, Florida, Indiana, Iowa, Louisiana, Mississippi, Nebraska, Ohio, Oklahoma, South Carolina, South Dakota, Tennessee, Texas, and Wyoming. Two of them signed a confidential memorandum that would let the department send back lists of voters to be removed within 45 days, and that permits sharing the data with contractors. Mississippi, South Dakota, and Tennessee refused to sign the agreement but provided the rolls anyway.³³

That memorandum is the part that should worry people who care about the rolls. The National Voter Registration Act requires notice and then two federal election cycles before a moved voter can be removed. A 45-day removal demand cannot be reconciled with that. And the contractor provision is a pipe from federal data to private organizations that file mass challenges.

The citizenship database. A March 2026 executive order directed DHS, with the Social Security Administration and the State Department, to compile and transmit “State Citizenship Lists” to every state’s chief election official, at least 60 days before each federal election. September 4 was the deadline for this November.³⁴

The underlying data is not fit for the purpose, and the government’s own documents say so. Social Security’s citizenship data reflects a person’s status when they applied for a number and does not update on naturalization; the agency did not consistently maintain citizenship information until 1981, so the data for anyone over 45 is incomplete. Social Security’s policy manual states that the agency “does not have the legal authority to disclose information about U.S. citizens to DHS.” USCIS guidance concedes that SAVE may not be able to confirm citizenship for people who acquired it through a parent. A DHS privacy analysis acknowledged the SAVE changes were “not in compliance with” the Privacy Act and suffered “shortfalls in data accuracy.” A ProPublica and Texas Tribune investigation found error rates as high as 81 percent in some localities.³⁵

The list of who gets flagged wrongly is not random. It is citizens born abroad, people over 45, people who changed their names, and people who move. That is the profile of a naturalized citizen, and it is also the profile of a married woman who took her husband’s name, which is why New Hampshire’s proof-of-citizenship rollout produced women unable to register without a marriage license.

The 250,000 number. In July, DHS announced it had identified more than 250,000 potential noncitizens on voter rolls across California, Nevada, New Jersey, and Pennsylvania. Nevada scaled its figure down from about 16,000 to 185. The department told Pennsylvania officials it “did not have high confidence” in the numbers. The letters used the construction “may be as many as.” By one tally, the department’s own letters matched 118,003 of 256,463 records, and they claimed Social Security number matches even though all four states’ public voter files lack Social Security numbers and all four are among the states that refused to hand over unredacted rolls.³⁶

The historical pattern is consistent: Florida’s 2012 list went from 180,000 to 85. Georgia found about 139 noncitizen votes cast since 1992, out of 32 million. Utah’s review found none. Louisiana looked back to the early 1980s and found 79 potential noncitizen voters out of more than 74 million ballots.

And the enforcement. ProPublica’s analysis of Justice Department data found fewer than 150 referrals, 41 people charged, and 14 convictions between January 2025 and May 2026, against more than 150 million people who voted in 2024. In Texas, 18 million registered voters produced 7 charges and 578 people wrongly flagged as noncitizens who turned out to be citizens.³⁷

None of this means the data campaign is harmless. It means the harm is not the fraud it claims to prevent. It is the wrongful flagging, the burden on election offices, the immigration consequences that follow from a bad match, and the foundation it lays for a contest after November.

Risk 7: Suppression by attrition

The quietest risks are the ones with no villain in the frame.

Election officials keep leaving. Turnover among local election officials reached 40.9 percent in 2024. More than half of chief local election officials have been in their roles less than five years. Half the counties in eleven Western states have lost their chief election official since November 2020. The pattern is not random: 80 percent of closely decided 2020 counties saw turnover, against 40 percent of blowout counties. The people who ran the hardest elections left. Pay is part of it, at roughly $50,000 against a $70,000 median for comparable public-sector roles, and sixty-hour weeks are normal in peak season.³⁸

The 2026 cycle has already produced departures weeks before the election: Chatham County, Georgia’s election supervisor resigned September 15; Anson County, North Carolina’s director left after three months; Chester County, Pennsylvania’s director resigned in March; a Nevada registrar cited threats to his family, with police checking his home hourly.

Threats are the reason, and they remain high. In the Brennan Center’s 2026 survey of 834 local officials, 32 percent reported being threatened, harassed, or abused. Of those who reported threats, 65 percent were threatened in person, most commonly at their offices. Nearly one in four worry about being assaulted. More than half worry threats will make it harder to recruit and retain workers. And 75 percent say their state or local government has not provided resources to offset the federal cuts.³⁹

Poll workers are the fragile layer. More than 772,000 people worked the 2024 election, and 48 percent of jurisdictions reported difficulty recruiting. Maricopa County was hiring more than 4,000 for this November in mid-September. Michigan’s Chamber of Commerce launched a statewide recruitment campaign in August. There is no national 2026 census, which is itself a gap.

Mass challenges are the cheapest suppression tool. At least 185,232 voter challenges were filed in 2024, likely more. Only about 3 percent were clearly accepted, meaning the whole enterprise contributed essentially nothing to keeping the rolls clean while consuming enormous staff time. Georgia saw 364,541 challenges filed in a single day, December 18, 2020, by one group, producing no removals at all. Two years later, roughly 89,000 of Georgia’s 92,000 challenges came from six people. Maricopa County saw 39 challenges covering 49,730 voters. In Gwinnett County, one organization filed 37,000 challenges; evaluating them consumed 5 to 10 experienced staff working six days a week, and removed not a single ineligible voter.⁴⁰

A federal court in Georgia described one such list as having “utterly lacked reliability… verges on recklessness… the list was shoddy and rife with errors.” A Republican member of Allegheny County’s board of elections said he could not count the hours he had wasted trying to debunk the challengers’ claims.

The new wrinkle is federal backing. The Justice Department’s data agreement allows sharing with any “contractor” performing list maintenance work, and the department reportedly met with the creator of the EagleAI challenge tool, whose account is “They like it.” The tool has since been relaunched under the name ELLY and pitched to North Carolina’s state board. Challenge laws have a specific history in this country: they were one of the devices states used to undermine the Fifteenth Amendment, and in 1946 a Georgia gubernatorial candidate distributed challenge forms urging supporters to disqualify three-quarters of the Black vote.⁴¹

The federal purge program is bigger than the number of people it catches. DHS has run the names of at least 67 million voters through government databases looking for noncitizens and people who have died: 60 million in a year, plus 7.4 million from North Carolina. USCIS identified about 24,000 potential noncitizens across those 60 million checks, and the assistant attorney general who runs the Civil Rights Division said in a Fox News interview that the same checks also turned up about 350,000 people who appear to have died.⁴² Even if every single flag were verified ineligible, noncitizens would amount to roughly 400 per million registrations. At least 25 states have used the system since April 2025, and six or more federal lawsuits have been filed over it.

Note the timing constraint: the National Voter Registration Act bars systematic removals within 90 days of a federal election, which put the blackout at about August 5 for this November. Ohio’s law requiring monthly SAVE and BMV checks was frozen by a judge in August, weeks before the registration deadline.

Eligibility rules are tightening. Five states now require documentary proof of citizenship from all registrants: Arizona, New Hampshire, South Dakota, Utah, and Wyoming. Seven voter-ID measures are on 2026 ballots, and Oklahoma voters added one to their constitution in August. Since January 2025, 19 states have enacted 44 restrictive voting laws, surpassing the previous record set in 2021 and 2022.⁴³

The population affected is not small. About 21.3 million voting-age American citizens, roughly 9 percent, do not have proof of citizenship readily available, and about 3.8 million have none at all. The disparity is racial: about 8 percent of white citizens and about 11 percent of citizens of color. Kansas’s similar law in the 2010s blocked more than 30,000 citizens from registering before it was struck down, against 39 confirmed noncitizen registrations in fourteen years.

And when polling places close, the effect is measurable. Tarrant County, Texas, which has 1.3 million registered voters and has been trending purple, voted 3 to 2 along party lines on September 1 to cut Election Day sites from 316 to 224. That is 92 fewer places to vote. One ZIP code would have had no site at all. The election administrator said the map concentrates “where voters choose to vote.” A commissioner called the reduction “modern day poll taxes.” After an August meeting on the plan, one attendee was carried out by deputies and another was arrested a week later.⁴⁴

Not every closure is sinister: Lubbock County closed two sites for a runoff because it could not staff them, and Karnes County consolidated from nine sites to eight after losing two veteran election judges. That is what a poll-worker shortage looks like when it reaches the map.

The mail-ballot rejection rate is the sharpest number here. California’s June primary rejected 148,427 mail ballots, or 1.73 percent of the 8.58 million returned, the highest rate in a decade. Nearly 93,700 of those rejections were for lateness. About 44,000 were for signature mismatch and 8,300 had no signature at all.⁴⁵

The lateness figure connects directly to the Postal Service’s postmark change. As the California Voter Foundation’s Kim Alexander put it, these ballots are rejected because they were “postmarked too late to count, not because they arrived too late to count.”⁴⁶ A rule that changes when a stamp gets applied, combined with a network redesign that means mail may not reach a processing facility the day it is mailed, produces rejections in states whose law says a ballot counts if it is postmarked by Election Day.

There is one genuinely bright finding here. A University of Pennsylvania study of the 2024 election found voters were 25 percentage points more likely to have their vote counted if they were told about a disqualifying error before Election Day, and 25 points more likely when deficient ballots were automatically returned or replaced so the cure could be done by mail. Delaware County’s director said automatic replacement roughly doubled the cure rate. The catch: about 20 percent of Pennsylvania mail-ballot requestors never supply an email, so they may never be told.⁴⁷

Ballot curing is the cheapest, least controversial fix available, and it works.

The protection apparatus was dismantled first

There is a version of this story that is about threats, and a more important version that is about the response to threats. The federal machinery built to protect election workers has been taken apart at the same time the threats peaked.

The DOJ Election Threats Task Force has no permanent leadership. The head of the FBI’s Elections Threats Task Force resigned in March 2025 and the role was unfilled as of April 2025. The Justice Department eliminated its centralized election-year command post, discontinued mandatory election-law training for prosecutors (not held since 2024), and restricted state officials’ access to threat briefings. The 24-hour FBI election-week rapid-response operation that fielded bomb-threat, hacking, and grid-failure calls is disbanded, with response dispersed to 93 U.S. attorneys, which former officials warn invites partisan inconsistency. CISA ended the Election Day situation room it had run for years.⁴⁸

One caveat, and it matters: no document confirms the task force was formally abolished, and none confirms it continues. Issue One describes its status as “currently uncertain.” The honest description is that its leadership is vacant and its supporting architecture is gone.

The task force’s own record before all this is instructive about how hard this is. Of more than 2,000 threats referred by election workers, the department opened about 100 investigations, charged 20 people, and won 15 convictions. Colorado’s secretary of state alone referred 1,140 threats since January 2023, producing 13 investigations and one prosecution. The reason conversion is so low is the “true threats” standard: prosecutors must prove a statement would reasonably be perceived as threatening, and the Supreme Court’s 2023 decision in Counterman v. Colorado added a subjective-awareness requirement. A vast majority of hostile messages are First Amendment-protected.

So the threat problem was never going to be solved by prosecution. But there is a difference between a low conviction rate and no one answering the phone.

The chilling effect is now measurable

Here is where the numbers stop being abstract. In the Brennan Center’s 2026 survey, 32 percent of local election officials reported being threatened, harassed, or abused. But the more revealing figure is what it did to them: 76 percent reported reducing at least one key civic activity because of hostility, up from 73 percent in the previous quarter and 66 percent two quarters before. Specifically, 43 percent are less likely to seek re-election, 47 percent are less likely to work on controversial topics, and 64 percent are less likely to use social media.⁴⁹

That is a ratchet. And 45 percent of threats go unreported, which means the survey is a floor.

The named departures are more eloquent than the rates. A California county elections official retired in 2024 saying she had been diagnosed with heart failure and that “an essential part of recovering… is stress reduction.” Her successor resigned for health reasons within a year. A Nevada registrar left after police began checking his home hourly. A Cochise County official described conditions as “physically and emotionally threatening”; that county then went through five elections directors in two and a half years. Pinal County, Arizona had four in four years. And three weeks before early voting opened this year, Chatham County, Georgia’s elections supervisor resigned six months into the job, citing board non-engagement and “backlash” after flagging wrongdoing.

Risk 8: The machinery has been hollowed out

Three structural facts about federal election infrastructure in 2026.

The Election Assistance Commission has no commissioners. Since July 9, when the president removed two Democrats and a Republican resigned, the agency that certifies voting equipment and maintains the voluntary standards has lacked a quorum. Career staff can still certify systems under existing rules and disburse appropriated funds. But the agency cannot adopt new standards, modify its manuals, change the federal voter registration form, or accredit and revoke testing labs. This is the third time the commission has lost its quorum. The White House cited a June 2026 Supreme Court case about removal powers at the Federal Trade Commission to justify the firings, an argument that assumes a president may fire members of a body Congress designed to be bipartisan.⁵⁰

CISA’s election security work has been dismantled. The Cybersecurity and Infrastructure Security Agency lost about a third of its workforce, from roughly 3,400 to about 2,500. Seventeen election-security staff were placed on leave in February 2025 and the program was paused for a review whose findings were never released. The FBI’s Foreign Influence Task Force was disbanded on the Attorney General’s first day. The agency did not stand up its Election Day situation room for the November 2025 elections, the first time in years. The current budget proposes eliminating election security entirely. No Senate-confirmed director has been seated since the administration began.⁵¹

The information-sharing center that election offices relied on, the EI-ISAC, was defunded in March 2025 and now runs on paid membership. It served about 3,700 jurisdictions in 2024. Rural offices say they cannot afford the fees. Congress appropriated $39.6 million for election security in April 2026; as of early September, DHS had not distributed it.⁵²

The trust damage is the part that may outlast the staffing. Arizona’s secretary of state deliberately did not tell CISA about a 2025 intrusion into his office’s websites, routing the notification through state officials without naming his office, saying, “I don’t trust them.” Michigan’s deputy secretary of state described the CISA relationships as destroyed, with local officials emailing and getting no response.

The equipment is aging and the standards are frozen. Most certified voting systems still trace to the 2002 or 2005 federal standards, and the commission that could update them cannot vote. In April 2026 the agency proposed retiring more than 45 older models. The vendor market also consolidated when Dominion was sold and rebranded as Liberty Vote.

What holds it together is the paper. More than 95 percent of registered voters are in jurisdictions with a verifiable paper record of their ballot, and fewer than 4 percent lack one. That is the backstop that makes the cyber risk survivable, and it is the single most important physical fact here.

Risk 9: Cyber, AI, and the information environment

On cyber, the honest answer is that nothing has gone badly wrong yet. No confirmed 2025 or 2026 ransomware or cyberattack has been shown to have touched vote tabulation or altered a vote total. The documented incidents are election-adjacent: Winona County, Minnesota, hit twice by ransomware in early 2026 and restored, with possible personal data exposure and credit monitoring offered; a municipal incident in Foster City, California; and the Arizona defacement. Voter registration databases are structurally separate from tabulation systems, are backed up offline and encrypted, and tend to fail visibly before Election Day because ballot requests and early voting act as integrity tests.

The historical record is worth carrying for context: Russia probed voter databases in all fifty states in 2016, Kennesaw State exposed about seven million Georgia voter records in 2017, and Iranian actors obtained voter data in at least one state in 2020. None of it changed a vote.

But the federal government’s own assessment is sobering, and it deserves careful handling because of the circumstances of its release. A CISA report declassified in July 2026, covering 2019 through 2024, found that in red-team exercises “CISA assessors gained full network control within hours or days” of state and local election networks. It documented flat networks, weak multifactor authentication, unpatched legacy software, and reliance on “airgap” assumptions that do not survive contact with vendor support connections. It also noted that certification rules can bar patches for months before an election. The report is a genuine agency product and it was also surfaced inside a contested political rollout. Both things are true. Its findings are corroborated; its timing is not innocent.⁵³

On AI, deepfakes are no longer an exotic threat. They are campaign furniture. The National Republican Senatorial Committee released an 85-second AI-generated “James Talarico” reading his own posts and inventing self-praising lines, with a disclosure label so small and faint it barely registered. A super PAC in Kentucky ran an AI ad depicting a congressman in a sexual scenario with two other members; the congressman lost and said it “was actually very effective on the Boomers.” There are unlabeled AI videos in Michigan and Oregon. A medical student in India fabricated a MAGA influencer who accumulated millions of views before being removed.⁵⁴

Thirty-one states now have election-deepfake laws. California’s and Hawaii’s were struck down on First Amendment grounds; Montana’s is being challenged; most exempt satire; and enforcement is spotty. The laws are being written faster than they can be tested.⁵⁵

The most interesting finding is about the chatbots. The Brennan Center tested ChatGPT, Gemini, Grok, Claude, Perplexity, and DeepSeek against six election conspiracy theories between February and August 2026. Every model disputed every theory, even under sustained pushback from a user insisting the system was rigged. That is genuinely reassuring. But roughly half of responses contained an inaccuracy or a bad citation, one in three contained a factual error, and one in three contained broken or misleading citations. Claude asserted that ballots are only tabulated when political party observers are present, which no state requires. Perplexity said the 2026 election had already happened.⁵⁶

So the tools resist the conspiracy and get the facts wrong. That is a strange failure mode, and it means the same systems that could help correct a rumor can also seed a new one.

On foreign interference, the signal is the absence of a signal. For the first time in about a decade, the Office of the Director of National Intelligence’s annual threat assessment omitted foreign election interference. Pressed at a Senate hearing in March, the director of national intelligence would not affirm a specific foreign threat but also would not deny one. Meanwhile the offices built to detect that threat are gone: the FBI task force, the ODNI center, the State Department’s engagement center. The Cyber Command and NSA chief testified in April that it is “reasonable to expect” foreign interference and that he did not know whether the joint election security group had been reconvened.⁵⁷

Risk 10: Enforcement aimed at people, not rules

Every mechanism discussed so far targets rules or data. This one targets individuals, and it is the part of the landscape with the fewest legal guardrails because most of it is ordinary prosecutorial discretion.

ICE is buying a national voter database. The agency announced a contract worth $2 to $5 million for “the handling and secure delivery of publicly available voter registration files and voter history files to support Homeland Security Investigations fraud detection and data segmentation activities.” The contractor must be able to obtain data from all fifty states, and ICE is seeking voting history going back across general, primary, runoff, and special federal elections.⁵⁸

The solicitation says the contractor will not be asked to collect party affiliation, but that it could change if “explicitly directed in writing by the contracting officer.” Separately, ICE was reported to be set to pay Thomson Reuters $125 million for access to its CLEAR database, which aggregates names, addresses, property records, driver’s licenses, social media profiles, and voter registration records. And in June, ICE agents obtained individual voter rolls directly from election officials in two counties in Texas and North Carolina, bypassing procurement entirely.⁵⁹

DHS directed ICE to deport noncitizens who vote, with no criminal conviction required. The June 9, 2026 directive from DHS General Counsel James Percival notes that the Immigration and Nationality Act’s removal provisions for illegal voting and false claims to citizenship “even allow for the removal of aliens legally in the U.S. if they illegally participate in our elections. A criminal conviction is not required to invoke these provisions.”⁶⁰

That is the sentence to sit with. Combine it with the SAVE database’s documented error rates, and the mechanism becomes: an erroneous match, a registration flag, and a civil immigration consequence that requires no conviction. This is what makes the data-quality problem matter to individuals rather than only to aggregate statistics. A derived citizen in Texas named in the EPIC lawsuit was already wrongly removed from the rolls once and reinstated only after proving his citizenship by hand.

197 election deniers are on the November ballot in 43 states. Twenty-seven already hold statewide office with election oversight (secretary of state or attorney general), and 33 more are running for such positions. Since 2022, States United Action has counted 525 such candidates.⁶¹

Trump’s own language has intensified: 60 percent of his statements referencing elections since 2024 contain an explicit claim of fraud, rigging, or theft. In the second quarter of 2026, that figure rose to 84 percent.

The races worth knowing:

• Nevada secretary of state. Jim Marchant, who calls “election denier” a “badge of honor” and helped organize a QAnon coalition of candidates in 2022. He has said he would deregister every voter in the state, move to paper ballots, and require all votes counted by hand. July filings showed $13.86 in his campaign account against more than $1 million for his opponent.
• Colorado secretary of state. James Wiley says 2020 was stolen and that his first act would be to delete Colorado’s voter lists and force every citizen to re-register, which is illegal under both federal and state law. He used the ELLY tool to claim up to 25 percent of Colorado registrations have problems.
• Wisconsin governor. Tom Tiffany signed onto the Texas lawsuit to invalidate Biden’s win and voted against certifying Pennsylvania and Arizona on January 6. As governor he would appoint members of the Wisconsin Elections Commission, the body that oversees voting for 2028.
• Ohio governor. Vivek Ramaswamy has called January 6 an “inside job” and “entrapment,” and supports enshrining restrictive voter ID in the state constitution.
• Michigan Senate. Mike Rogers, who in January 2021 called Trump’s Georgia call “more gangster than presidential” and said of January 6, “You’re damn right you had responsibility for this,” thanked Trump for “commitment to transparency” moments after a July 2026 fraud speech.

Wendy Weiser of the Brennan Center frames the stakes: “All Americans should be concerned if people who are trying to undermine trust in elections and trying to interfere in the election process are taking positions of responsibility over elections… This really is an attempt to corrode the system from within.”

The Wisconsin case is the structurally interesting one. The state’s secretary of state does not currently oversee elections; that power sits with the bipartisan Elections Commission. The Republican nominee for that office campaigns explicitly to move election authority back to the partisan office. Wisconsin is one of the states that will certify the 2028 presidential result.

Risk 11: The precedent that shapes all of it

One thing in this landscape is not a forecast. It already happened, and it tells you what the environment looks like now.

On January 20, 2025, the president granted clemency to roughly 1,600 people convicted of or awaiting trial for the January 6 attack, commuting 14 Oath Keepers and Proud Boys sentences and issuing “full, complete, and unconditional” pardons to the rest. The Justice Department was ordered to dismiss pending indictments with prejudice. In November 2025, he pardoned 77 “alternate electors,” including Giuliani, Powell, Eastman, Ellis, Chesebro, Meadows, and Clark.⁶²

The recidivism data is the part that has gone underreported. By June 2026, Lawfare had counted at least 97 clemency recipients subsequently arrested, charged, or convicted, roughly one in sixteen, including at least 14 charged with sex crimes or child sexual abuse material, six with domestic violence, and more than 20 with DUI or public intoxication.

The most striking reaction came from law enforcement itself. The Fraternal Order of Police, which had endorsed Trump three times, and the International Association of Chiefs of Police jointly said that pardoning “individuals convicted of killing or assaulting law enforcement officers… sends a dangerous message that the consequences for attacking law enforcement are not severe.” Eighty-three percent of Americans opposed pardoning violent participants.

And the institutional cost: the department fired two supervising attorneys and a line attorney who had handled January 6 cases. The attorney who led the Capitol Siege section before it was disbanded resigned saying the pardons “send… that political violence towards a political goal is acceptable in a modern democratic society.”

Why this belongs in a piece about November 2026: the clearest available signal about how seriously a subversion attempt would be punished is how the last one was. Whatever one thinks of the individual pardons, the message sent to anyone considering intimidating a poll worker or refusing to certify a result is the one the police unions identified, that the consequences are not severe.

What is actually holding

It is worth being as clear about the defenses as about the risks, because the defenses are the reason this election is likely to be run well.

The Elections Clause does the heavy lifting. Every adverse ruling in this entire landscape rests on one sentence: the times, places, and manner of holding elections “shall be prescribed in each State by the Legislature thereof,” with Congress able to override by statute. The president has no role. That is not a technicality. It is the reason the executive orders died, the reason the Postal Service rule died, the reason the voter-roll suits died, and the reason the citizenship database is in court.

State and local officials of both parties have refused. The Pennsylvania secretary of the commonwealth is a Republican who said the federal numbers on noncitizens were unreliable. A Nevada Republican official said the state did not want to accuse a citizen of not being a citizen and leave them terrified and disenfranchised. Twenty-four courts have agreed with the states. The system held because dozens of appointed officials chose to follow the law, most of them unknown outside their counties.

Paper ballots. More than 95 percent of voters have a paper trail. Machines can be wrong; paper can be recounted.

Courts have been fast. Courts have procedures for ruling quickly on election disputes, and they have used them. Mandamus actions have compelled certification in five states. The Supreme Court denied the Postal Service rule nine days before the overseas-ballot deadline, which is fast for anything.

And voters have the last move. Turnout is the one variable nobody controls. Every mechanism described here, from mass challenges to agent presence to a delayed certification, depends on people deciding the system is not worth the trouble. The most robust protection is not a court order. It is not needing one.

What to watch between now and January 6

1. The FEMA grant deadline and which states refuse the conditions. This is the live mechanism that no court has touched.
2. Certification in the seven battleground states, on the dates above, and specifically whether any county refuses and how fast the remedy lands. The number to watch is not whether it happens but how many days it costs.
3. The Maryland case on the citizenship lists, and whether the lists are actually transmitted to any state. As of September 16, Pennsylvania had still not received its list despite the department promising it a week earlier.
4. Whether the Eighth Circuit revives Missouri’s HB 1 map, which currently is not in force.
5. The Justice Department’s monitor deployment, which jurisdictions, and whether any monitors are removed for violating state law. Monitors’ conduct has already been questioned in Wyoming, where they sought access to voting machines, and in Minnesota, where a monitor had a prior conviction for hacking state election websites.
6. Election night arithmetic in states that cannot pre-process mail ballots. Pennsylvania and Wisconsin are two of only six states that cannot begin processing mail ballots until Election Day, which means they count slowly by design. The gap between early returns and the final count is the single most exploited feature of modern American elections, and it is a feature of state law, not a malfunction.⁶³
7. The mail-ballot rejection rate in November. California’s June primary rejected 1.73 percent of mail ballots, the highest in a decade, and nearly 93,700 of those were for lateness. If that rate holds nationally in a higher-turnout general election, it is the largest single source of disenfranchisement in the country, and almost none of it involves fraud.
8. The secretary of state and governor races in Nevada, Colorado, Wisconsin, and Ohio, where the winners will run the 2028 presidential election.
9. Congress on January 6, 2027, and seating the week before. The electoral count that day applies only to presidential elections, so 2026 is a congressional certification and seating question under Article I, Section 5, not an electoral-count question. Watch January 3, when the 120th Congress convenes, and whether any member-elect is challenged, asked to stand aside, or refused the oath. A single challenge would be the first documented step toward the scenario in Risk 4; the absence of one is what keeps that risk ranked where it is. The machinery built after 2020 gets its first real test in 2028, and the officials elected in 2026 will be the ones operating it.

The honest uncertainty

Four things I cannot tell you, and neither can anyone else.

Whether any of this changes turnout, and in which direction. Both Gallup and UC San Diego note the effect is genuinely unknown. Concern can deter people, and it can also motivate them. There is no 2026 data linking the claims to a measured turnout effect. This is the single largest hole in the whole picture, and it is the one that decides everything else.

Whether Iran actually attempts to interfere. The claim that major technology companies gave a confidential briefing asserting a “high degree of confidence” that Iran will attempt to disrupt the general election comes from a single reported account. The 2026 threat assessment omits the category entirely. The honest label is unknown.⁶⁴

What the Supreme Court will do next. The Postal Service rule was denied on timing, not authority. Justice Kavanaugh wrote that there is “at least a fair prospect” the rule falls within the Postal Service’s statutory authority. The underlying question is unresolved, and it will come back.

Whether anything has been prepared for the aftermath. I could not find a single sourced account of specific Capitol, statehouse, or certification-security planning for November’s aftermath or the January 2027 count. The federal rapid-response structure that would have coordinated it has been disbanded. The nearest evidence is the 2024 precedent of National Guard standby in three states and twelve-hour police shifts in Washington. This is a genuine gap in the public record, and it may mean the planning is quiet rather than absent. It should not be filled with inference.

The thing worth holding onto

The 2020 and 2022 and 2024 elections were all preceded by predictions that the machinery would break. It did not break. It bent, in specific places, and the places are known now, and most of them have been reinforced.

The defenses that held are unglamorous and worth naming plainly. A constitutional clause that assigns elections to states and Congress and gives the president nothing. Thousands of county officials, of both parties, who did their jobs. Paper. Courts that moved fast when asked. And the simple fact that the country has now been through this four times and the system has not yet produced a wrong winner.

What is different this time is not the machinery. It is the ambient belief about whether the machinery is worth operating. The rules have survived the last five years remarkably well. The harder question is whether the country still agrees that the rules are the point.

There is one more thing worth saying, and it is the least speculative claim in this piece. The most likely thing to go wrong on November 3 is not a stolen election or a seized ballot box or a federal agent at a polling place. It is that the votes get counted correctly, that the results get certified on time, that the losing side has a legitimate grievance about close races decided by small margins, and that a large share of the country declines to accept any of it.

That outcome requires nothing to be broken. It only requires enough people to stop believing the count. Which means the defense is not legal. It is the same defense it has always been: show up, and then believe what the paper says.

Notes

PRH | huffmanwrites.org | © Philip Huffman