Objection (high): Tense and modality overclaim in the headline claim. The PSA says actors "may attempt to disrupt the 2026 U.S. midterm elections" and that it is describing "techniques... that malicious cyber actors have used during past election cycles." The claim converts this into a present-indicative assertion: actors "are using phishing emails and smishing messages ... to steal login credentials and sensitive data from election and campaign officials ahead of the 2026 U.S. midterm elections." That asserts an ongoing, currently-observed campaign against 2026 targets. The source supports (a) a historical surge over the last two cycles, (b) an FBI-observed rise in smishing via encrypted chat apps in the last year, and (c) a forward-looking warning. It does not support "are using ... ahead of the 2026 midterms" as a statement of present fact. The finding body gets this right ("may try to disrupt"); the claim line does not. Fix the claim to modal/attributed phrasing, e.g. "warn that actors may target..." or "say election officials have faced a surge of phishing and a recent rise in smishing."
Not resolved — preserved on the record.
Objection (high): Causal fusion between AI/SI and the observed phishing. The claim welds two separate source statements into one causal assertion: that the messages hitting election officials are "increasingly polished with AI-style tools." The PSA never says any observed election-targeting campaign was SI-generated. It says, separately, that campaigns have grown sophisticated "in recent years," and that "the rise in SI creates additional opportunities" and "SI allows" actors to write natural language. Those are capability/opportunity statements, not attribution of observed messages to SI tooling. The alternative explanation — that the polish is from human-written, well-resourced state or criminal operators, translation services, or stolen legitimate templates — is not excluded by anything in the row. The deepfake sentence is likewise generic ("adversaries and criminals"), with no claim that deepfakes have been used against election targets; the draft's own unknowns list concedes this, which contradicts the confidence of the headline.
Not resolved — preserved on the record.
Objection (medium): Terminology substitution: the source says "super intelligence (SI)" throughout and never says "AI," "artificial intelligence," or "AI-style." The finding handles this correctly by quoting "super intelligence" and attributing the term to the advisory, but the claim silently relabels it "AI-style tools." That is the checker interpreting an agency's idiosyncratic term and shipping the interpretation as the top-line claim. Either carry the agency's word in the claim or mark the gloss explicitly. It also matters as a source-integrity signal: "super intelligence" is unusual phrasing for a joint FBI/CISA PSA and is worth a second look before it is paraphrased into something more familiar-sounding.
Resolved: Partly resolved already: the finding sentence quotes "super intelligence" and attributes the term to the advisory rather than asserting it in the checker's own voice. Residual fix needed only in the claim line.
Objection (medium): The single most prominent mitigation in the source is dropped. The PSA's first tip to individuals and a repeated tip to organizations is "implement phishing-resistant MFA," with a linked CISA fact sheet, plus "examine potentially malicious content for content credentials or watermarking indicating that it was SI-generated." The draft's advice sentence covers only second-channel verification. For an alert explicitly addressed to election and campaign officials, omitting the agencies' headline control while keeping the softer one misrepresents the advisory's emphasis.
Not resolved — preserved on the record.
Objection (medium): Date currency / shelf life is unverified. The row is dated 2026-10-16 and the 2026 federal midterm election day falls in early November 2026. "Ahead of the 2026 U.S. midterm elections" is only true if this alert publishes before election day. Nothing in the draft records the publication date of the alert relative to the election, and no freshness check is noted. If this ships after election day the framing is stale and the urgency is fabricated.
Not resolved — preserved on the record.
Objection (medium): Single-source dependency on a URL that is not independently corroborated. Everything rests on one ic3.gov PSA page. The claim attributes the warning jointly to FBI and CISA, but no CISA-side posting or mirror is cited, and no independent reporting is cited. The draft discloses the single-source limitation, which is good, but the entity/domain check is incomplete: a joint PSA attributed to two agencies should be confirmed on at least one of the two agencies' own domains beyond the IC3 PSA page before the claim names CISA as a co-author.
Not resolved — preserved on the record.
Objection (low): Attribution precision drift. The draft says "The FBI reports that election officials saw a surge of phishing attempts during the last two election cycles." In the source, the surge sentence is unattributed within the joint PSA ("During the last two election cycles, election officials experienced a surge of phishing attempts"); only the smishing clause is attributed to the FBI ("the FBI saw an increase"). Attributing both halves to the FBI adds a specificity the text does not carry. Say "the advisory says" for the first clause.
Not resolved — preserved on the record.
Objection (low): Audience scoping is narrower than the source. The PSA repeatedly addresses "election officials and the American public" and explicitly says it "provides the public with information." The callout "ATTENTION: ELECTION AND CAMPAIGN OFFICIALS" is defensible as the primary audience, but the finding should not leave readers with the impression that the advisory is officials-only; the finding does mention the public once, so this is a presentation nit rather than an evidence error.
Resolved: Largely resolved by the finding sentence "The agencies urge election officials and the American public to critically evaluate the sources of their emails and mobile messages," which preserves the dual audience.
Objection (low): risk_line mild dramatization: "can hand over your login and your organisation's data in one click." The source says phishing may lead to disclosure of credentials or financial information, or malware installation. It does not establish a one-click path from message to organisational data loss. Also, "organisation" is non-US spelling in an alert about US federal agencies and US elections — cosmetic, but it reads as a template artefact.
Not resolved — preserved on the record.
Objection (low): Internal inconsistency in the suppression record. targeting_dropped reads "DIRECTIVE NOT IN EVIDENCE: working — no harvested source describes this group," but none of the four directive_options mentions a "working" group; two of them are incoherent ("Send this to any multi-factor you know"). The suppression note does not correspond to the candidates it is supposedly explaining. The outcome (share_directive null) is correct, but the audit trail is wrong and would not survive review. Same category: callout_options includes "ATTENTION: VETERANS" and "ATTENTION: SCAMMEDS", neither of which has any basis in the row — these should not be generated at all for this evidence set.
Resolved: Outcome is acceptable — share_directive is null and the sole chosen callout is supported by the advisory's named target group. Only the explanatory note needs correcting; no public-facing text is affected.
Preserved dissentON THE RECORDThe claim as written is an overclaim and I would not sign it. The advisory says actors "may attempt" to disrupt the 2026 midterms and that it is describing techniques used "during past election cycles." The claim says actors "are using" these techniques against election and campaign officials ahead of the midterms. That is a modal-to-indicative upgrade on the single most consequential word in the sentence, and the finding body contradicts the claim line by correctly saying "may try." Rewrite the claim or drop it.
ON THE RECORDI also dissent on the AI framing. Nothing in the harvested row attributes a single observed message, campaign, or deepfake to SI tooling. The PSA makes a capability argument ("SI allows," "creates additional opportunities"), and the draft converts it into an observational one ("increasingly polished with AI-style tools"). The draft's own unknowns list admits we do not know whether deepfakes have been used against election targets at all. Putting the AI angle in the headline is the most clickable reading of the source, not the best-supported one.
ON THE RECORD"Super intelligence" is the advisory's word, not "AI." The checker should not quietly normalise an agency's unusual terminology into the familiar term in the top-line claim. If the agencies wrote something odd, report it as odd.
ON THE RECORDModerate confidence is, if anything, generous. Source reliability is high and the quotations are accurate, but the substantive proposition the alert is selling — active targeting of 2026 election officials — is supported by zero incident evidence, zero counts, zero senders, zero samples. I would cap this at moderate only if the claim is rewritten to the modal form; with the current claim line I would call it low.