FBI and CISA: AI-written phishing and deepfakes target US election officials before the midterms

On October 16, 2026, the FBI and the Cybersecurity and Infrastructure Security Agency (CISA) warned that attackers are using AI-written phishing messages and deepfake audio and video against election and campaign officials ahead of the November 3 US midterms. The public service announcement (PSA) names no new malware and publishes no indicators; its value is a clear, short defensive checklist — phishing-resistant MFA first — that applies to any organisation, not only election offices.

What happened, in plain English

The FBI's Internet Crime Complaint Center (IC3) published PSA I-101626-PSA, co-signed with CISA, the US agency responsible for protecting government networks and critical infrastructure. It warns that election officials, candidates and political organisations are being targeted by phishing (fake emails) and smishing (fake text messages) designed to steal passwords and other sensitive data, or to install malware.

Two things are new compared with earlier election cycles. First, generative AI — which the PSA calls "super intelligence (SI)" — lets attackers write convincing messages in a target's native language and everyday slang, and run larger campaigns that improve over time. The old advice to "look for bad grammar" no longer works. Second, criminals use AI to produce deepfakes: realistic fake audio or video of real people, used to create fake identities, impersonate someone the victim trusts, or make a phishing request more believable.

The agencies also note that election officials saw a surge of phishing attempts during the last two election cycles, and that within the past year the FBI saw more smishing through encrypted chat applications.

An analogy: the con artist used to send a badly forged letter. Now they can send a perfectly written one — and follow up with a phone call in your boss's voice. The defence is no longer spotting the forgery; it is having a process where even a perfect forgery cannot get the keys.

Reuters, via The Star, reported that the alert was issued on Friday, October 16, a little over two weeks before election day.

Are you affected? What to do now

Directly affected: staff at US state and county election offices, campaigns, party organisations, and anyone who supports them — IT contractors, managed service providers, election-technology vendors, and the helpdesks of local governments.

Indirectly relevant to everyone else: nothing in this PSA is election-specific in its mechanics. AI-polished phishing and voice-cloned "urgent requests" hit finance, HR and IT helpdesks everywhere. If you are outside this sector, use the list below as a quick self-check; there is no patch to apply and no emergency action required.

Checklist, in priority order (based on the PSA's recommendations):

• Phishing-resistant MFA on email, VPN and critical systems. This means FIDO2 security keys, passkeys or smart cards — methods that cannot be relayed by a fake login page. SMS codes and push-approve prompts are better than nothing but can be phished. CISA's fact sheet on phishing-resistant MFA explains the options.
• Out-of-band verification for unusual requests. Any request for credentials, money, voter data or access changes is confirmed through a second channel using contact details you already have — not the number or link in the message, and not a reply in the same chat thread. This is the main defence against deepfake voice and video.
• A one-click way to report suspicious messages to IT, and a clear instruction not to open attachments from unknown senders.
• Email authentication and filtering: publish SPF (Sender Policy Framework) and DKIM (DomainKeys Identified Mail) records to make spoofing your domain harder; add external-sender banners; enable strong spam filtering; block emails carrying executable files.
• Strong, unique passphrases for remaining password-based accounts — especially service, admin and domain admin accounts — and rotate them after any compromise.
• Least privilege: role-based access, reviewed regularly, so one stolen account cannot reach everything.
• Patching: keep systems updated and watch CISA's Known Exploited Vulnerabilities catalog.
• Awareness refresher before November 3: tell staff that well-written messages and familiar voices are no longer proof of authenticity; cover text and encrypted-chat lures, not just email. The PSA also suggests checking suspicious media for content credentials or watermarks indicating AI generation — useful when present, but their absence proves nothing.
• Document incidents and their remediation clearly.

Report suspicious activity to your local FBI field office or to IC3. The PSA publishes no indicators of compromise (IOCs).

The expert view

There is little technically new here, and that is the point. The PSA describes no novel exploit, no named threat actor and no attribution to any country. What it does is formally register, for a high-stakes sector with a fixed deadline, that the cost of producing a credible, personalised lure has collapsed.

Conceptually, AI changes phishing on two axes. The first is quality at scale: models remove the linguistic tells that filters and trained users relied on, and allow cheap per-recipient tailoring. Threat-intel teams have documented this repeatedly — for example, Cisco Talos recently described a spear-phishing campaign against Taiwan-based researchers whose emails showed signs of a reusable AI prompt template, paired with a kit that relays real Google sign-in and MFA steps in real time. The second is channel expansion: voice and video, once the trusted fallback for verification, can now be synthesised. A process that says "if in doubt, call them" is only safe if you call a number you already hold.

This is why the PSA leads with phishing-resistant MFA. Most AI-assisted credential phishing still ends at a login page. Adversary-in-the-middle kits forward whatever the victim types — password, SMS code, even push approval — to the real service. FIDO2 and passkeys bind the authentication to the genuine website's domain, so a lookalike page receives nothing reusable, however convincing the email that led there. Better writing does not defeat cryptographic origin binding.

The shift to encrypted chat apps matters for defenders' visibility. Those messages bypass corporate email gateways entirely, so detection depends on people reporting them and on policies that say official requests never arrive that way.

What remains unknown: the PSA gives no numbers on successful compromises, no detail on which AI tools were used, and no confirmed case of a deepfake used against an election office in this cycle. Its terminology — "super intelligence" for today's generative AI — is unusual and should not be read as a claim about a new class of capability. The practical takeaway is unchanged from good security hygiene: make the stolen password worthless, and make verification independent of the channel the attacker controls.

Official sources

• FBI / IC3 and CISA — Cyber Threat Actors Use a Variety of Techniques to Target Election and Campaign Officials (I-101626-PSA), October 16, 2026
• CISA — Implementing Phishing-Resistant MFA (fact sheet)
• CISA, FBI and partners — Phishing Guidance: Stopping the Attack Cycle at Phase One
• CISA — 2026 Election Infrastructure Security Plan
• FBI — Election crimes and security
• Cisco Talos — UAT-11985: AI-assisted event lures delivering real-time Google AitM phishing (context)
• Coverage: Reuters via The Star — US tells election officials to be on guard against AI-powered tricks ahead of midterms