The Tool Screening Your Sewing-Operator Applicants Is Regulated. The Camera Watching for Skipped Stitches Isn't.

"We cut fleece in Gastonia, not software." Why August 2 reaches you anyway

This is the objection I hear in every plant, and it is a fair one. You run knitting machines and a cutting room. You have never trained a model, you do not sell software, and the idea that a European regulation has anything to say about your operation sounds like consultant talk. Mostly, you are right. But not entirely, and the exceptions are specific enough to check in an afternoon.

Three dates have already landed. The EU AI Act's obligations for high-risk uses and for transparency carry a compliance date of August 2, 2026, and they reach US companies whose systems affect people in the EU — which for this trade usually means a brand customer in Europe or a webshop that ships there. Texas TRAIGA and California SB 53 both took effect January 1, 2026. Colorado, New York, Utah, Nevada, Maine and Illinois all have their own AI statutes. Federal preemption of state rules is still unsettled as of this July, so state law binds you today regardless of what Congress does later.

For a US apparel manufacturer, almost none of this is about the machines on your floor — it is about software that decides something about a person, or that talks to a person as though it were staff.

The three things in a plant that actually attract the rules

Go down your list of software and you will find that exactly three categories matter, and they are probably not the ones you expected.

• Hiring and workforce tools. If anything screens, ranks, scores or schedules interviews for sewing operator, cutter, presser or supervisor applicants, that is a decision about a person's employment. This is the heaviest-regulated category by a distance: it is high-risk under the EU rules, it is squarely covered by the Illinois amendments that took effect this January, New York City has required bias audits of automated employment tools since 2023, and Colorado's law covers consequential decisions including employment. Plants with heavy operator turnover — which is most of them — are the ones most likely to have bought an automated screening tool without thinking of it as AI.
• Anything that talks to a person as if it were a person. The chat window on your B2B site, the voice agent answering the sales line, an automated reply to a wholesale inquiry. The obligation is disclosure: the person on the other end must be able to tell they are dealing with a machine.
• Anything scoring people other than employees. Credit-like decisions on a small wholesale account, for instance. Rare in a contract shop, common in a brand with a direct wholesale book.

What you concretely have to document, disclose or log

The practical answer is smaller than the anxiety around it. For each tool in those three categories you need, on paper: what it is and who supplies it; what decision it influences and who makes the final call; what data goes into it; how a person can ask for a human review; and the fact that a named employee is responsible for it. For the hiring tool specifically, add the notice you give applicants and, in New York City, the published bias audit.

If a customer-facing chat or voice agent is in use, add the disclosure wording and keep transcripts. Keep them the way you keep anything else that might be asked for in an audit — retrievable, dated, and not on somebody's personal laptop.

flowchart TD
  A["List every AI tool touching the plant"] --> B{"Does it decide who gets hired, scheduled or promoted?"}
  B -->|Yes| C["High risk: log it, keep human review, notify the applicant"]
  B -->|No| D{"Does it talk to a person as if it were staff?"}
  D -->|Yes| E["Disclose it is a machine and keep transcripts"]
  D -->|No| F{"Do EU buyers or EU shoppers see its output?"}
  F -->|Yes| G["Add it to the register your EU brand customer will ask for"]
  F -->|No| H["Out of scope: write down the decision and date it"]

What is genuinely out of scope, and you can stop worrying about it

Here is the part nobody selling compliance services will tell you. Most of the AI in a modern textile plant is not regulated by any of this.

The marker-nesting optimiser in Gerber AccuMark or Lectra that squeezes another 1.4 percent out of your fabric utilisation: out of scope. It decides about cloth, not people. The camera on the inspection table that flags skipped stitches, needle holes or a shade variation between rolls: out of scope. The demand forecast that sets your size curve: out of scope. Predictive maintenance on a Shima Seiki flat knitting machine or a Tajima embroidery head: out of scope. Automated fabric defect grading on the perch: out of scope.

California SB 53 deserves a specific mention because it comes up constantly and is widely misread. It is aimed at the companies building the very largest models — the ones with enormous training budgets and public safety obligations. You buy software; you do not build frontier models. It does not apply to your plant. Texas TRAIGA is likewise focused on prohibited uses and on state agency use, with a narrower reach into ordinary private commerce than the headlines suggested.

Writing down why something is out of scope is worth as much as documenting what is in scope. When a brand's vendor compliance team sends you a questionnaire, "we assessed it and here is our reasoning, dated March 2026" is a complete answer. Silence is not.

The Tuesday version: two pages in the binder you already keep

You already maintain a compliance binder. WRAP certification audit file, SMETA report, OEKO-TEX or GOTS certificates if you carry them, CPSC Children's Product Certificates and the tracking label records if you make kids' apparel, flammability test records under 16 CFR 1610, fibre content and care labelling records under the FTC rules, and your forced-labour traceability documentation for CBP. Adding a two-page AI register to that binder is a smaller job than any single one of those.

Practically: your HR manager lists the hiring tools, your marketing or sales lead lists anything that chats with customers, your IT contact lists anything else, and one person — usually the controller or the compliance manager who already owns the WRAP file — signs the page. Review it every six months, on the same calendar as your other audit prep. If you sell to an EU brand, expect their vendor questionnaire to ask for exactly this within the next year.

The arithmetic on doing it versus not

The costs here are not the penalties, which vary and which nobody should quote at you casually. The real arithmetic is the account.

Set that against one EU brand programme worth, say, $1.9 million a year in cut-make-trim revenue, held up while their compliance team waits for a questionnaire answer you cannot produce. The math is not close. This is a paperwork afternoon protecting a customer relationship, which is the same calculation you already made about WRAP.

Where you need a lawyer, not a checklist

Three situations warrant a real employment or trade attorney and not an article. First, if you use any automated screening on hourly applicants and you operate in Illinois, New York City, Colorado or California, the interaction between those statutes and existing discrimination law is genuinely unsettled and the exposure is employment litigation, not regulatory fines. Second, if you sell direct to consumers in the EU under your own label, your scope question is broader than a contract shop's and worth a proper opinion. Third, if a brand customer sends you contract language making you responsible for their AI compliance, read it carefully — indemnity clauses have been quietly appearing in vendor agreements this year, and signing one moves somebody else's risk onto your balance sheet.

Everything else on the list above is genuinely a two-page exercise your own team can do.

Frequently asked questions

We are a contract sewer. Our customer is the brand. Isn't this their problem?

For their products and their marketing, yes. For your hiring, your phone line and your own chat window, no — those are yours regardless of who owns the label in the garment. The split is simple: whoever operates the tool carries the obligation for how it is used.

Does the fabric inspection camera need to be registered?

No. It examines cloth, not people, and it is not making a decision about anyone's rights or employment. Write one line in the register saying you reviewed it and concluded it is out of scope, with the date, and move on. That line is worth more than it looks when a questionnaire arrives.

We ship to a distributor in Rotterdam who sells our goods on. Are we in scope?

Selling goods to Europe does not put you in scope by itself — the trigger is a software system of yours affecting people there. If your website chat answers European shoppers, or you screen applicants at an EU location, look closer. If your only European contact is a container and an invoice, you are almost certainly out. Ask counsel if a real programme depends on the answer.

Our applicant tracking system has an AI matching feature we never turned on. Does it count?

Check whether it is genuinely off, get that in writing from the vendor, and record it. A surprising number of plants have scoring features switched on by default in their applicant tracking system and would tell an auditor honestly that they do not use AI in hiring. That is the single most common gap I see.

A note on where CallSphere fits. If you put a voice or chat agent on your sales line, disclosure and record-keeping are part of doing it properly, not an afterthought. CallSphere builds AI voice and chat agents that answer business phone lines and web chat, book appointments and capture leads around the clock — with the disclosure that the caller is speaking to an assistant, and a written transcript of every conversation you can hand to an auditor or a brand's vendor compliance team.