Safety training in a language workers understand: what the rule requires, and what it leaves open

OSHA's Training Standards Policy Statement of 28 April 2010 states that employee training required by OSHA standards "must be presented in a language and vocabulary that employees can understand." It applies across every OSHA standard carrying a training requirement. It is not guidance, and it does not scale with how many workers are affected.

Most safety leaders know that. The part worth ten minutes of your morning is what the sentence leaves out.

Which direction does the language requirement run?

Read it again and watch which way the information moves.

Training goes out. Hazard communication goes out. So does the induction, the safety data sheet, the pre-job briefing, the toolbox talk. All of it is the company sending something to the worker, and the rule says the company has to send it in a form he can take in.

Now look for the equivalent coming back. The hazard he wants to raise. The near miss on the next bench. The observation card. The pre-task assessment he has to complete before the crew starts.

There is no equivalent.

We brief a man in Spanish and hand him a form in English. Both halves of that sentence are compliant.

ISO 45001 gets closer than OSHA does. Clause 7.4 asks that communication processes account for diversity, and language and literacy sit inside that. Clause 5.4 asks for worker consultation and participation, which is hard to demonstrate when a third of the crew cannot use the channel. Neither clause is written as a hard language requirement on the reporting path, and neither gets audited that way.

So the gap is not a loophole anybody exploited. Nobody built the second half.

Why doesn't translating the form fix it?

The obvious fix is to translate the form. Plenty of operators have, and it helps. It does not finish the job, for two reasons that get run together.

Language and literacy are separate constraints. A man can speak fluently and still not write well, in his own language or in yours. Translating a fifteen-field form into Portuguese gives a Portuguese-speaking worker with limited literacy a fifteen-field form he still cannot complete. He does what he has always done, which is find a supervisor and tell him. The supervisor writes up his version of what he was told, and what reaches the system is second-hand.

A form is a translation exercise before it is a language problem. Even in his first language, the worker has to convert what happened into the categories the system holds. Which of eleven hazard types. Which body part. Which severity band. Which of four root causes, none of which describe it. That conversion is work, it is unpaid, and it happens at the end of a shift when he wants to go home. Translating the field labels leaves it untouched.

There is a third effect, and it shows up in the record rather than in the participation rate. Detail dies in translation. A man describing a leaking tote in his own words gives you the smell, the fitting it came from, and that it is the second time this month. The same man working through a dropdown gives you "chemical spill, minor". Both are true. One of them is useful to whoever reads it next. We went through that in more detail in where safety reports lose the detail that mattered.

What does the workforce actually look like?

The US Bureau of Labor Statistics recorded 5,070 fatal work injuries in 2024, a rate of 3.3 per 100,000 full-time equivalent workers. Among Hispanic or Latino workers there were 1,229, at a rate of 4.3, and 68.5 per cent of those fatalities were among foreign-born workers.

That is workforce composition and exposure. It belongs in front of you when you decide which language your reporting path runs in, and it is not an argument about software.

The reporting side has its own figures. A 2025 study in Occupational and Environmental Medicine linked Illinois hospital records to OSHA's severe-injury reports across 2017 to 2023 and found a cumulative reporting rate of 39.7 per cent for non-fatal injuries. In Great Britain, HSE's published position is that employer reporting of RIDDOR-defined non-fatal injuries to employees "is estimated at around a half."

Those cover injuries serious enough to be legally reportable. Whatever the equivalent rate is for hazards and near misses, where nothing compels anybody, it is not higher.

Five things to check on your own sites this week

None of these need a vendor. Most can be answered from data you already hold.

1. What languages are spoken across your sites, and what language are your forms written in? Ask site managers rather than HR. HR has a nationality field. Site managers know who needs a translator in a briefing.
2. What share of last quarter's hazard reports were submitted by a worker, and what share by a supervisor on a worker's behalf? If your system does not distinguish those, that is the finding.
3. Pull twenty reports at random from your highest-hazard site and read the free text. Count how many run past two lines. Count how many read like somebody was there.
4. Compare reporting rates between your most and least linguistically mixed site. If the mixed site reports less per hundred workers, you have located the problem without having to prove the mechanism.
5. Ask a supervisor how long he spends each week writing up other people's accounts. Nobody measures this and every one of them has a number.

If four and five come back clean, the language path is not your constraint and you can stop reading vendors on the subject. That is a useful result and it costs an afternoon.

Can a worker report in his own language and still produce a structured record?

Yes. The two are not in tension.

Tandm lets a worker report a hazard, an incident or a pre-task assessment by speaking or typing in his own language, at the job. Tandm works out what is missing from the account he is giving, asks him for it, and shows him the completed version before anything is recorded in his name. He never opens a form. The structured record still exists on the company's side, in the language and shape the system needs.

The record goes into the safety system already in place. Nothing is replaced and nothing is migrated.

We are not claiming this fixes a reporting culture. Plenty of things suppress reporting that have nothing to do with language, and a crew that believes raising a hazard gets a job stopped and a bonus lost will stay quiet in any language you offer. What it removes is the mechanical part: a man who saw something, could describe it in ninety seconds, and was handed a screen instead.

Book a demo and pick the language yourself. Twenty minutes, live, in whichever language you like. It is a faster read on whether this works than anything we could write.

Sources

• Occupational Safety and Health Administration, Training Standards Policy Statement, 28 April 2010. osha.gov/training/compliance
• US Bureau of Labor Statistics, National Census of Fatal Occupational Injuries in 2024, released 19 February 2026.
• Friedman LS, Forst L, Shannon B, Abasilim C, Madigan D. Employer compliance with OSHA requirements for immediate reporting of severe injuries. Occupational and Environmental Medicine 2025;82(3):148–156.
• Health and Safety Executive (Great Britain), Non-fatal injuries at work in Great Britain, 2024/25.
• ISO 45001:2018, clauses 5.4 and 7.4.