Nearly 200 Class Actions Over Four-Word Subject Lines

TV
Thiago Victorino
8 min read
Nearly 200 Class Actions Over Four-Word Subject Lines

“Ends tonight” is now worth $500 per email, per recipient, in Washington State. Nearly 200 class actions have been filed under the state’s Commercial Electronic Mail Act against retailers, airlines, hotels, colleges, insurers and beauty brands. Costco settled for $14 million, with preliminary approval on June 23, 2026 and claims due August 24, 2026. In the reporting on these filings, not a single plaintiff claims to have actually read the challenged emails.

What follows is an operational read of a live litigation trend, not legal advice. The reason it belongs on an AI governance desk rather than a legal one is the arithmetic. CEMA prices liability per send, and the volume of sends is exactly what an agentic marketing stack was bought to increase.

Damages that attach at the moment of send

CEMA has been in force in Washington since 1998. It prohibits commercial email containing false or misleading information in the subject line, and every violation is automatically an unfair or deceptive practice under the state Consumer Protection Act. That routing matters more than the statute itself: it carries attorney fees and the possibility of trebled damages. Before this year’s amendment, statutory damages ran $500 per email, per recipient, with no cap and no requirement to prove harm. Receiving the email is the injury.

For years the assumption was that this reached sender spoofing. Brown v. Old Navy, certified to the Washington Supreme Court on April 17, 2025, closed that reading. Any false or misleading information in a subject line violates the law, not only deception about who sent the message. Urgency claims came inside the perimeter.

The filings that followed read like a copy deck. Ma v. Nike, filed in King County in May 2025 and later in the Western District of Washington, targets “Only a few hours left”, “Ends tonight” and “Last chance to save an extra 25%”. Nike’s preemption defense failed on January 14, 2026, because CAN-SPAM, when Congress enacted preemption in 2003, expressly spared state statutes prohibiting falsity or deception. Haley v. Tommy Bahama went after “Ends Tonight!” and “Limited Time Only”. Macy’s drew a suit over “Open ASAP: up to 65% off ends tomorrow”. Costco’s settlement covers subject lines including “Today is the last day to access Member-Only Saving” and “Hot Buys available for 5 Days Only”.

Standing comes from a private right of action for Washington residents, a population of eight million. The statute reaches a sender who has reason to know a recipient lives there, and plaintiffs meet that standard with the signup data already sitting in your own platform. Douglas Karr, writing for MarTech Zone, puts exposure for a single national campaign touching a few hundred thousand Washington inboxes in the hundreds of millions of dollars. No court has awarded damages at trial in any of these cases yet, so settlements are where the real penalties are currently being set.

The generation rate is the multiplier

Every governance framework I have seen for AI marketing assumes the risk is qualitative: a hallucinated claim, an off-brand tone, a fabricated statistic. Someone catches it, someone escalates, the campaign gets fixed. CEMA does not work that way. It assigns a fixed number to each individual message that left the building, and the machine’s whole advantage is that a lot of messages leave the building.

Karr names the operational surface precisely: evergreen journeys, resend-to-non-opener flows and recycled templates repeat final hours language on autopilot. Audit what the system sends, not only what the calendar plans. A campaign brief that specifies a real deadline can still produce a violation six weeks later when the journey fires the same template at a new segment and the deadline has long passed. The human who wrote the brief is gone from the loop by then. The send is not.

That is the shape of the problem I described in Marketing Just Got Its First Engineering Outage, now with a price tag attached to each unreviewed output. Variant generation makes it worse in a specific way. Ten subject line variants tested across a list is ten times the surface, and the variants that win an A/B test are the ones that lean hardest on urgency, which is to say the ones most likely to overstate a deadline the operations team never actually set.

Email is the forensic surface

Dela Quist, founder of Alchemy Worx, made the point in a December 2025 LinkedIn post quoted by MarTech Zone: email is being singled out because it is traceable, testable, and codified. He calls it the forensic surface of marketing. TV implies urgency. Shelf wobblers suggest urgency. Email documents it.

Karr’s version is blunter. A sale sign that stays up an extra day is unprovable; an email is its own receipt, metadata included. Which leads to the sentence that should reorder anyone’s priorities here: your exposure is not prospective, it is sitting in Sent folders. The archive is held by the plaintiff class, timestamped, with headers intact, and no retention policy of yours touches it.

After June 11, your logs are the discovery target

House Bill 2274, signed by Governor Bob Ferguson on March 23, 2026 and effective June 11, 2026, cut statutory damages to $100 per email per recipient and added a knowledge requirement. Plaintiffs must now show actual knowledge, or knowledge fairly implied from the circumstances, that the subject line was false or misleading when it was sent. The amendment is not retroactive, so the roughly 200 cases filed before June 11 proceed under the old $500 framework. Karr’s read is that the legislature installed a speed bump, not a roadblock.

The second-order effect is the one worth planning around. A knowledge standard turns discovery toward the record of what the operator knew at send time. In a stack where a human wrote each subject line, that record is thin. In an agentic stack it is enormous and it is automatic: the prompt, the campaign configuration, the promotion table the generator read from, the model output log, the approval that was or was not requested. “Knowledge fairly implied from the circumstances” is a standard that a well-instrumented pipeline satisfies against its own operator. If the offer record shows no end date and the generator produced “Ends tonight”, the circumstances documented themselves.

This is the operator-carries-the-output principle from A Court Told Google It Owns Its AI’s Mistakes meeting the discoverability problem from Log Retention Is a Legal Instrument. Both were abstract a year ago. Washington made them a line item.

The control the plaintiffs’ bar is pricing

Strip the litigation down and one control is being tested: a named human who checks each urgency claim against operational reality before the message goes out. That is precisely the gate an agentic marketing stack is designed to remove, and its absence is now priced per send.

The audit Karr proposes is six questions, and none of them are legal questions. Does the deadline exist? Is the discount real? Does the subject line match the offer? What is your automation saying? Can you prove it, meaning do you retain records of start dates, end dates and terms? Who signs off? An engineering team would recognize every one of these as a precondition check, and would have wired them into the pipeline rather than a checklist.

There is a commercial argument underneath the legal one. Manufactured urgency is a depreciating asset. Audiences learn quickly that your deadlines are decorative, and real deadlines honestly enforced are what make the tactic work at all. The compliance control and the performance control are the same control, which is the position I argued in Marketing AI Now Has a Reputational Bill and a Disclosure Clock.

Do this now

Export every subject line your automation sent to Washington recipients since June 2021 and join it against your promotions table. You are looking for sends where the urgency language has no matching end date in the offer record, and the largest cluster will be in evergreen journeys and resend flows rather than in anything a marketer scheduled by hand. That set is your pre-amendment exposure and it is already fixed in size, so measure it before deciding anything else.

Then put the gate upstream. Any generated subject line containing a deadline claim resolves against a structured offer record with a real end timestamp, or it does not send. Log the resolution. Under a knowledge standard, the log that proves the check ran is worth more than the copy it approved.


This analysis synthesizes Ends Tonight? How False Urgency Became Email’s $500 per Subscriber Mistake (Douglas Karr, MarTech Zone, August 2026).

Victorino Group helps marketing and engineering teams build the send-time controls their AI content pipelines now require. Let’s talk.

All articles on The Thinking Wire are written with the assistance of Anthropic's Opus LLM. Each piece goes through multi-agent research to verify facts and surface contradictions, followed by human review and approval before publication. If you find any inaccurate information or wish to contact our editorial team, please reach out at editorial@victorinollc.com . About The Thinking Wire →

If this resonates, let's talk

We help companies implement AI without losing control.

Schedule a Conversation