EU Bans Deepfakes in August. US Votes in November.

On August 2, 2026, the EU's deepfake labelling law kicks in. Every AI-generated video, image, or audio clip must carry a machine-readable tag marking it as synthetic. Companies that skip the label face fines up to €35 million — or 7% of global revenue, whichever is larger. Ninety-two days later, Americans vote in the midterms.

That gap isn't a coincidence. It's the story.

Two democracies, one technology, opposite rules

The EU's Article 50 comes from the AI Act passed in 2024. The transparency obligations were always scheduled for August 2026 — the law gave companies two years to comply. AI deployers must disclose when synthetic content is artificially generated. That applies to everyone serving EU markets: Meta, Google, TikTok, OpenAI, all of them.

The US has no equivalent. Federal law has no statute specifically regulating AI-generated political deepfakes. It has older laws instead — fraud, election interference, identity theft, defamation — all written before generative AI existed.

Twenty-six states have passed some form of deepfake election law. Most require disclosure or ban deceptive synthetic media within a set window before election day. But they contradict each other: some cover only video, some exempt satire, some only apply in the 60 days before an election. None are federal. None cover primaries.

The other 24 states have nothing.

The war tested this first

The Iran conflict ran the experiment before the midterms.

Iran's IRGC deployed deepfakes within hours of its opening strikes on February 28 — including videos falsely claiming to have killed Israeli PM Benjamin Netanyahu. Within seconds, a follow-up message purporting to be Israeli authorities directed people to a malicious app. A real strike, a disinformation push, and a cyberattack — coordinated in a way analysts hadn't seen at scale before.

Pakistan joined in. Pakistan-linked accounts circulated AI-generated videos of India's Chief of Naval Staff, Admiral Dinesh K. Tripathi, and External Affairs Minister S. Jaishankar — both flagged as deepfakes by India's PIB Fact Check unit.

The Foreign Affairs Forum assessed that AI-generated content now constitutes a larger share of the disinformation ecosystem in the Iran conflict than traditionally manipulated material. A qualitative threshold, not just more of the same.

The same companies, the same tools, the same models used in that conflict will be serving US political ads in October.

The first confirmed midterm deepfake

On March 11, the National Republican Senatorial Committee posted an attack ad featuring a deepfake of James Talarico, Democratic nominee for US Senate in Texas. Talarico called it illegal. His legal team is checking which state laws might apply. The NRSC hasn't taken it down.

Texas has a deepfake disclosure law. Whether it applies to this ad depends on timing and whether the content meets the statutory definition — being debated by lawyers, not regulators.

That's the practical gap. Even where state laws exist, enforcement requires lawsuits. The FEC hasn't issued rules on AI campaign content. Congress hasn't passed a bill. The primary runoff is late May.

The EU law starts August 2. The general election is November 4.

What Article 50 actually requires

The EU rule covers AI systems generating synthetic audio, image, video, or text. Providers must mark outputs in a machine-readable format — effective, interoperable, robust, reliable. Deployers must disclose when content is artificially generated or manipulated, unless it's obvious, artistic, or authorised for law enforcement.

There's a satire exemption. It's already being flagged as a loophole — campaigns could argue attack ads are "satirical" commentary. EU regulators are aware. The Commission's Code of Practice, published in January 2026, doesn't resolve it cleanly.

Article 50 is a transparency obligation, not a prohibited practice. Fines run up to €15 million or 3% of global revenue. The 7% figure applies to the most serious violations under Article 5.

A Meta or Google deepfake noncompliance fine could still run to hundreds of millions. In the US, the equivalent fine is zero.

The structural divergence

Deepfakes in elections aren't new — that's been a concern since 2019. What's new is the timing collision.

The EU enforcement deadline lands three months before the US general election. The same AI tools, built by the same American companies, will face binding legal obligations on one side of the Atlantic and effectively none on the other. A political AI company in Berlin must label its content. The same company's US division can put the same video into a Texas attack ad with no federal requirement to disclose anything.

EU-compliant tools are more expensive — labelling infrastructure costs money. US campaigns using identical tools face no equivalent cost. That's not just a policy gap. It's a structural advantage for synthetic political content in the one election the EU law doesn't reach.

The Iran war has already shown what unlabelled AI content does to an information environment. The deepfake of Netanyahu's death circulated for hours before debunking caught up. Pakistan's deepfaked Indian military leaders reached audiences in multiple countries before anyone flagged them.

The 2026 midterms won't be a combat zone. But the same tools, at higher volume, aimed at 240 million eligible American voters, with no federal labelling rule — that's not a hypothetical. It's what the calendar already shows.

The EU built a warning label system. The US will run the November election to find out if it needed one.

---

For more on how information shapes perception across borders, see the Albis Perception Gap Index and our AI and information warfare coverage.