What 43 safety technology vendors publish for buyers: the 2026 evidence dataset

Before a sales call, a buyer can usually learn who a safety technology vendor is and how its product is installed. What they usually cannot learn is how accurate the product is, what it costs or how it handles workers' data. That is the main finding from checking the public pages of 43 vendors against the same twelve items on 3 and 4 October 2026 ^[1].

The full results are published as an open dataset, free to reuse with credit ^[2][3][5]. This article summarizes what it shows and how to use it.

What we checked

Every vendor in our directory was checked against the same twelve items: a security or trust page, a published data processing agreement (DPA), an ISO 27001 or SOC 2 claim, documented privacy features, documented deployment architecture, public pricing or a pricing model, a published accuracy or validation method, independent evidence, a named customer case study, listed integrations, company details, and at least five third-party reviews ^[1]. An item counts only if a public page showed it on the day we checked, and the profile links to that page ^[4].

The 43 vendors cover AI video safety analytics (14), wearables and sensors (9), EHS software (9), video platforms (6) and other safety technology such as proximity and forklift systems (5) ^[2].

Five findings

Accuracy methods are almost never public. Only 3 of 43 vendors publish how they measure detection accuracy or validate their models ^[2]. Many more publish accuracy figures or customer results, but without the method a buyer cannot tell what was counted, on which footage, or over what period. Ask for precision and recall per detection type, measured in conditions like yours, before a pilot is agreed.

Pricing is mostly hidden. 13 of 43 vendors show prices or a pricing model in public ^[2]. The rest require a sales conversation, which makes early budget comparisons hard. Ask for the full three-year cost, including hardware, installation and any per-camera or per-user fees.

Data protection documents are uneven. 33 vendors claim ISO 27001 or SOC 2 and 26 have a security or trust page, but only 18 publish a data processing agreement ^[2]. For any system that films or measures workers, ask for the DPA, the list of sub-processors and the data retention settings in the first round.

Independent evidence exists for just over half. 25 of 43 vendors have at least one piece of evidence not produced by the vendor itself, such as a peer-reviewed study, an analyst report or a public purchasing contract ^[2]. Independent evidence was rarer among wearables vendors (4 of 9) than among EHS software vendors (8 of 9).

Reviews are hard to verify. Third-party review sites blocked our automated checks for 17 vendors, so that item is the least complete in the dataset ^[2]. Where we could check, 15 vendors had at least five reviews on a third-party platform.

Differences by category

EHS software vendors publish the most, with a median of 10 of the 12 items, followed by video platforms (8). AI video safety analytics and wearables vendors publish a median of 6, and the other safety technology vendors a median of 5 ^[2]. The biggest gaps between categories are in data processing agreements (6 of 9 EHS software vendors publish one, 4 of 14 AI video vendors and 1 of 9 wearables vendors) and security pages (8 of 9, 8 of 14 and 3 of 9) ^[2].

Each item, and what to request when it is missing

The table lists every item in the order of how often vendors publish it ^[2]. The last column is the document or answer to request from any vendor that does not publish the item.

The same items by category

Categories differ in size, so compare the share within each category rather than the raw counts ^[2].

How the checks were done

Each vendor's own website was read first: product, security, legal, pricing, customer and documentation pages. We then searched for independent sources such as peer-reviewed studies, analyst reports, public purchasing records and regulator documents, and for review listings on third-party platforms ^[1]. An item was recorded as found only with a link to the page that shows it. Certification and outcome claims were recorded as the vendor's claims; we did not verify certificates with the issuing bodies. When a site blocked automated access, we retried through an archived copy where one existed, and otherwise marked the item as not checked rather than not found ^[1][2].

Each vendor's profile in the directory shows its own checklist with the source for every item, and the dataset records the date each vendor was checked ^[2][3].

How to use the data

• Plan your first-round questions. Each missing item for a vendor on your shortlist is a document to request. The buying and piloting chapter explains how to turn them into an RFP.
• Compare vendors side by side. The vendor directory can filter vendors by the items you need, and the comparison table shows every item for every vendor.
• Reuse the dataset. Download the CSV or JSON from the data page. Each row is one vendor, with the status of every item and the link to the page where we found it ^[2][3].

Limits

This is desk research, not a test of products. A vendor with few public items may have an excellent product and share its documents under NDA, and a vendor with many items may still be a poor fit for your site ^[1]. The checks reflect public pages on two days in October 2026. Vendors can send corrections with a public link, and we recheck them for the next edition.