The Catastrophic Risks of Inadequate Signature Verification in U.S. Elections – A Five-Alarm Fire for Democracy

October 19, 2025

by Mark Cook

Executive Summary

Signature verification for mail-in ballots is a failed critical safeguard in U.S. elections, purporting to balance voter access with fraud prevention. However, reliance on Automatic Signature Verification (ASV) systems—plagued by high error rates in low-exemplar, low-resolution conditions—or fallback to untrained election workers (with minimal PDF-based or 1–2 hour training) creates an untenable vulnerability. Every rejected ballot is a silenced voice, every accepted fake a diluted republic. This report synthesizes real-world data to expose the dual threats: massive disenfranchisement of legitimate voters via False Rejection Rates (FRR), and unchecked fraud via False Acceptance Rates (FAR) from forged ballots. In the 2024 election cycle alone, over 122,000 California ballots were rejected due to signature issues—scaling to 1–2 million nationwide—disproportionately affecting young, new, and minority voters. Worst-case scenarios could disenfranchise millions or inject hundreds of thousands of fraudulent votes, rendering election outcomes unverifiable and untrustworthy. This is not mere inefficiency—it’s a systemic crisis demanding immediate reform, as unverified signatures erode the foundational security of American democracy.

Key Differences Between FAR and FRR

FAR (False Acceptance Rate) and FRR (False Rejection Rate) are common terms in security systems like fingerprint scanners, password checks, or signature verification for mail ballots. Think of them as the two main ways a system can “mess up.” Here’s a straightforward breakdown of how they differ, explained like you’re dealing with a bouncer at a club:

Aspect FAR (False Acceptance Rate) FRR (False Rejection Rate)
What It Means The system wrongly lets in an imposter—like a fake ID sneaking past the bouncer. (E.g., a forged signature gets approved as genuine.) The system wrongly kicks out a legit person—like your real ID gets rejected because it’s a tiny bit smudged. (E.g., your own signature gets flagged as suspicious.)
Real-World Risk Security threat: Bad guys win. In elections, this could mean fraudulent votes slipping through, messing with results. Convenience/injustice issue: Good guys lose. In elections, this could mean valid voters get their ballots tossed, silencing their voice.
Why It Happens The system is too “trusting” or lenient—set to avoid rejecting real people, but it overdoes it and accepts fakes too easily. The system is too “strict” or picky—set to block fakes, but it overdoes it and blocks real ones too.
How to Fix It Tighten the rules (e.g., demand more proof), but this might accidentally block some real users. Loosen the rules a bit (e.g., allow more wiggle room for variations), but this risks letting in more fakes.
Trade-Off Example In a bank ATM: FAR might let a thief withdraw cash; better to err on the safe side here. In a busy airport scanner: FRR might delay honest travelers; better to prioritize speed and let most through.

In short, FAR is about unwanted intruders getting in (big security no-no), while FRR is about turning away friends (big fairness no-no). Real systems try to balance both, but it’s a constant tug-of-war—when one goes up, the other goes down.

Inherent Dangers of Relying on ASV in Elections

ASV systems, while efficient for high-volume screening, falter dramatically in election-specific conditions: single or few low-resolution exemplars (e.g., 150–300 DPI scans from voter registrations), handwriting variations due to age/disability, and the absence of dynamic data (e.g., pressure/velocity). As detailed in prior analyses, standalone offline ASV yields worst-case FAR of 20–42% and FRR of 15–46% in single-exemplar tests with skilled forgeries and noise—rates that balloon disenfranchisement and enable fraud without human intervention.

In elections, ASV is often deployed as a first-pass filter, auto-approving “matches” (exposing FAR risks) and flagging mismatches for review (amplifying FRR if review is bypassed or inadequate). Real-world pilots, such as those in Colorado and California using tools like Parascript, show aggregate rejection rates of 0.5–3%, but these mask standalone ASV contributions: a 2020 Stanford study found ASV alone increased rejections by 74% compared to manual processes, implying FRR spikes of 10–20% pre-review. For fraud, targeted forgeries (e.g., by organized actors) exploit FAR laxity—systems tuned for low FRR to avoid backlash can accept 30–40% of skilled fakes, per BiosecurID and MCYT datasets.

The unmitigated danger: ASV lacks explainability, cannot adapt to contextual nuances (e.g., cultural scripts or tremors), and propagates biases, rejecting 2–5x more ballots from Black, Latino, elderly, and disabled voters. Without certified oversight, ASV becomes a black box for insecurity.

Perils of Non-Certified Human Verifiers

Election workers tasked with “curing” flagged ballots receive woefully inadequate preparation: self-study via PDFs (e.g., state guidelines like Arizona’s 10-page manual) or cursory 1–2 hour sessions from vendors/professionals. This leaves them wholly unqualified, akin to laypeople in forensic contexts, where accuracy plummets.

Forensic studies confirm untrained individuals achieve only 20–50% accuracy in handwriting comparisons, with error rates 6x higher than experts (who hit 93–95% on non-disguised samples). A 2019 pilot on novices showed near-perfect sourcing of natural signatures but rampant errors (up to 70% false positives/negatives) on disguised or variable ones—mirroring election ballots with aged or stressed writing. Large-scale validity research (e.g., NIST human factors studies) reveals lay verifiers exhibit bias, fatigue, and inconsistency, misclassifying 40–60% of cases under time pressure, especially for marginalized groups whose signatures deviate from “norms.”

In practice, this manifests as arbitrary rejections: A 2024 Maricopa County analysis found flaws in verification processes disproportionately burdening young and new voters, with rejection rates 2–3x higher for first-time mailers. Untrained workers cannot discern subtle forgeries or valid variations, inverting safeguards—FRR soars to 30–50% (disenfranchising voters) while FAR lingers at 15–30% (admitting fakes). Unlike certified forensic document examiners (requiring 2,000+ hours training), these ad-hoc verifiers lack peer review or error-tracking, amplifying systemic unreliability.

Potential Percentages of Voter Disenfranchisement from Forged Ballots and Verification Failures

Bad actors forging mail-in signatures (e.g., via stolen voter data or insider access) exploit FAR vulnerabilities, while verification errors drive FRR-based disenfranchisement. Using 2024 turnout data (~158 million total votes, ~40% mail-in or 63.2 million ballots, per U.S. Election Assistance Commission projections), we model worst-case impacts from ASV/untrained human rates. Assumptions: 1% of mail ballots targeted for forgery (realistic for coordinated efforts, per CISA risk assessments); no guaranteed review (per state variability).

Scenario Key Error Driver Disenfranchisement Mechanism Potential Impact (% of Mail Ballots Affected) Absolute Votes at Risk (2024)
High FRR (Legitimate Voters Rejected) ASV Standalone (15–46%) or Untrained Human (30–50%) Valid signatures flagged/rejected due to variability/low-res exemplars; no cure access. 15–50% of all mail ballots (skewed 2–5x for minorities/elderly). 9.5–31.6 million disenfranchised (e.g., CA’s 122k rejections scaled nationally = ~0.8–1.5% baseline, but worst-case triples to 3%).
High FAR (Forged Ballots Accepted) ASV (20–42%) on Targeted Forgeries; Untrained Oversight (15–30% miss rate). 1% forged ballots (632k) submitted; 20–42% accepted without scrutiny. 0.2–0.42% of total votes fraudulent (up to 1% in swing districts). 126–265k illegitimate votes counted; could flip 5–10 House seats or 0.1–0.3% national margin.
Combined Crisis (Hybrid Failure) ASV flags + Untrained Review (40–60% total error). 70% auto-approve (FAR exposure) + 30% manual mismatches (FRR); fraud + rejection cascade. 20–40% total ballots compromised (disenfranchisement + fraud). 12.6–25.3 million affected; erodes trust in 10–20% of results.

These figures draw from 2020–2024 data: California rejected 0.8% of 2024 mail ballots (122k total), up from 100k+ across 2020–2022, with racial disparities (e.g., 2x for Latinos). Nationally, signature mismatches caused 0.5–2% rejections in mail-heavy states, but untrained errors could inflate to 5–10% per 2024 studies on worker calibration. Forgery risks, though rare historically, amplify: A single coordinated effort (e.g., 0.1% ballots) at 30% FAR yields 19k fraudulent votes—enough to sway close races like Georgia 2020 (11k margin).

The Five-Alarm Fire: A Cataclysmic Threat to Election Security

This is no contained risk—it’s a raging inferno engulfing our Constitutional Republic. Unverified signatures create an unverifiable black hole: ASV’s opacity hides errors, while untrained humans inject subjectivity without accountability, yielding results neither auditable nor defensible. In 2024, amid polarized trust (only 58% confidence in elections per Gallup), a 1–3% error margin could delegitimize outcomes, fueling challenges like 2020’s 60+ lawsuits. Disenfranchisement hits hardest at the margins—young (2–4x rejection rates), disabled (up to 5x), and people of color—exacerbating inequities and suppressing turnout by 5–10% in affected demographics.

Fraud vectors compound the blaze: Forged ballots, once accepted, evade post-election audits (signatures aren’t traced), enabling undetectable dilution of votes. CISA warns of “insider threats” in mail processing, where FAR laxity invites tampering. With no federal standards for training or ASV validation, states patchwork solutions—some auto-reject without notice—turning safeguards into saboteurs. This un-verified, un-verifiable system invites exploitation: Bad actors need only 0.01–0.1% penetration to contest results, as seen in hypothetical models flipping 5–15 electoral votes.

The alarm blares: Without certified experts, we cannot rely on signatures as a secure gatekeeper. It’s a dereliction of duty, torching voter confidence and inviting chaos. There is nothing more valuable to a United States Citizen than their vote, and it would be criminal to put that vote at risk because of ASV.

Federal Testing Requirements for Signature Verifiers in U.S. Elections: Ensuring Zero-Tolerance for False Acceptances

To safeguard election integrity and prevent disenfranchisement through fraudulent vote acceptance, Congress must enact comprehensive, mandatory testing protocols for both Automatic Signature Verification (ASV) systems and human verifiers. Current federal law, such as the Help America Vote Act (HAVA) of 2002, mandates provisional ballots and voter verification but lacks specific proficiency standards for signature matching—leaving it to states with patchwork approaches (e.g., 34 states require signature verification for mail ballots, but only a few outline training or testing). This gap exposes vulnerabilities: ASV error rates can exceed 20% FAR in low-exemplar scenarios, while untrained humans achieve only 20–50% accuracy. Forensic standards from bodies like the European Network of Forensic Science Institutes (ENFSI) and ANSI/ASB emphasize annual external proficiency tests, but U.S. elections need tailored, enforceable rules.

Requiring 0% observed FAR (no forged signatures accepted as genuine) on a blind test set is a stringent but essential benchmark, mirroring forensic ideals where error rates near 0% are targeted for high-stakes conclusions. However, a minimum of 100 signatures is insufficient for statistical confidence—statistical models (e.g., Clopper-Pearson intervals) show that with 0 errors on 100 forgery samples, the 95% upper confidence bound on true FAR is ~3%, meaning the real rate could be as high as 3% undetected. To bound true FAR below 0.5% with 95% confidence, at least 500 known forgeries (plus 500 genuines for balance) are needed, for a total deck of 1,000 samples (upper bound ~0.5%). Congress should mandate 500+ known forgeries per test, scaled for diversity (e.g., cultural scripts, tremors).

Proposed Legislative Framework: Amendments to HAVA or New Election Security Act

Congress should require states to certify verifiers annually via federally accredited labs (e.g., NIST-overseen), with non-compliance triggering federal funding cuts. Tests must be blind, externally administered, and include diverse signatures (e.g., elderly, non-Latin scripts, tremors) to mitigate biases. Below is a table outlining minimum requirements, drawing from forensic standards (e.g., SWGDOC, ENFSI) adapted for elections.

Requirement Category ASV Systems Human Verifiers Rationale & Minimum Threshold
Frequency Annual recertification + post-update testing (e.g., after software patches). Annual proficiency testing + initial certification (e.g., 40-hour forensic-equivalent training). Aligns with ENFSI’s “at least one external test/year” to detect degradation; NIST recommends ongoing human factors validation.
Test Structure Blind set: 500 known forgeries + 500 genuine exemplars (1:1 ratio); low-res (150–300 DPI), single-exemplar matches. Include 20% disguised/simulated forgeries. Blind set: Same as ASV, presented sequentially to simulate workload; no time limits but audited for bias/fatigue. Forensic tests use 10–20 items but scale up for elections’ volume; 1,000 total ensures balance (FAR/FRR).
FAR Threshold 0% observed (0/500 forgeries accepted); upper bound <0.5% at 95% confidence. 0% observed; require “definitely forged” conclusion on all. Zero-tolerance prevents fraud; 500 samples needed vs. 100 (which only bounds <3%).
FRR Threshold <5% (≤25/500 genuine rejected); adjustable via cure process. <5%; allow “inconclusive” but not on forgeries. Balances access; NIST notes inconclusives as non-errors but FAR must be absolute.
Diversity & Conditions 30% non-Latin scripts, 20% with tremors/aging effects; variable scans (e.g., envelopes). Same; train on cultural/tremor recognition. Addresses biases (2–5x higher errors for minorities/elderly); per NIST human factors.
Certification & Oversight Vendor submission to NIST/FBI lab; open-source algorithms preferred. CFDE-equivalent (e.g., SAFE certification) + state licensing; peer review on 10% of tests. Ensures validity; ANSI/ASB requires documented skills.
Reporting & Penalties Public dashboards on error bounds; decertify if >0% FAR. Individual decertification + retraining; aggregate state reports to EAC. Transparency per NIST roadmap; ties to HAVA funding.

Enacting these via HAVA amendments would professionalize verification, reducing disenfranchisement (e.g., California’s 122,000+ rejections in 2024) while fortifying against fraud. This isn’t overkill—it’s the bare minimum for trust in democracy.

Comprehensive List of References and Links

Below is a compiled list of all key references, studies, reports, and data sources mentioned across our discussion on automatic signature verification (ASV), error rates, and election security risks. I’ve organized them thematically for clarity, drawing from peer-reviewed papers, databases, government reports, and analyses. Where direct links to full texts (e.g., PDFs) or official pages are available from recent searches (as of October 19, 2025), I’ve included them. Some older studies may require academic access (e.g., via Google Scholar or ResearchGate). If a reference lacks a precise match in search results, I’ve noted the closest verified source or DOI for retrieval.

1. ASV Databases and Competitions

2. Key ASV Studies and Surveys

3. Election-Specific Reports and Analyses