A Texas employer that opts out of workers' compensation makes a specific, rational trade. It gives up guaranteed premium costs and gains flexibility and savings. But it also gives up something less visible until the day it matters: the legal shield that keeps an injured worker from suing in civil court.
Texas is the only state that lets most private employers decline workers' comp entirely. Employers who do — "non-subscribers" — forfeit the exclusive-remedy protection that subscribers enjoy. They also lose the common-law defenses that normally blunt a negligence suit: contributory negligence, assumption of risk, and the fellow-servant rule. In plain terms, an injured employee can take a non-subscriber to court, and the employer walks in stripped of its usual armor.
Lower, more flexible insurance cost in exchange for direct, uncapped civil exposure — where one serious incident, argued as negligence, can cost more than years of premium savings combined.
None of this is a reason not to go non-subscriber. Many well-run companies do, and do it responsibly, because the savings are real and their safety culture is genuine. The point is narrower and sharper: once you're a non-subscriber, your ability to prove you acted reasonably is no longer a compliance nicety. It's your defense. And that is exactly where AI quietly changes the math.
The safety system you bought to lower risk
To manage that exposure, non-subscribers increasingly deploy AI safety systems: computer-vision cameras that flag missing PPE, sensors that detect unsafe proximity to equipment, analytics that surface near-misses. The pitch is intuitive — catch hazards in real time, prevent injuries, lower your risk.
Prevention is real, and a system that genuinely reduces injuries is worth having. But here is the part the vendor demo doesn't cover: an AI safety system is also a recording device. Every flag it raises is a timestamped data point. Every hazard it detects becomes a document. And documents are discoverable.
The same system that proves you were watching can prove you saw the danger and did nothing about it. Which one it proves depends entirely on what happened after the flag.
Proof or exposure: the fork
Consider two non-subscribers running the identical camera system. A worker is injured. Both get subpoenaed. Their data tells opposite stories.
| The system generates… | For Employer A (proof) | For Employer B (exposure) |
|---|---|---|
| A flag: unsafe condition detected, 9:14 a.m. | Reviewed, logged, corrected by 9:31 a.m., closed with a note. | One of 4,000 flags nobody reviewed. No response recorded. |
| A pattern: same hazard flagged 12 times in a month | Escalated, root-caused, fixed — documented improvement. | A documented record that the risk was known and repeated. |
| Threshold settings on the detection model | Chosen and signed off by a named safety lead, with rationale. | Vendor defaults. Nobody can say why they're set where they are. |
Employer A's system is the best evidence in the courtroom that the company was diligent. Employer B's system is the plaintiff's strongest exhibit — a machine-generated timeline proving the hazard was seen, logged, and ignored. Same tool. Opposite outcome. The difference isn't the AI. It's the governance around it.
Buying a safety AI without governing it doesn't just fail to reduce your non-subscriber exposure. It can increase it — by manufacturing, at scale, the exact evidence of "known and ignored" that turns an ordinary claim into a gross-negligence argument.
What "due diligence" actually has to look like
Because the non-subscriber's whole defense is "we acted reasonably," the AI safety system has to be able to demonstrate reasonable action — not just detection. That means four things a plaintiff's attorney will look for, and that you want to already have:
- Human-in-the-loop: When the system flags a hazard, is there a documented human review-and-response protocol — or does it flag into a void?
- Closed loops: Can you show each significant flag was addressed and the loop closed — or do open flags sit in the record as unhandled known risks?
- Design defensibility: Can a named person credibly explain why thresholds and camera placements were set as they were, as a reasonable safety decision?
- Reliability: Is anyone monitoring the false-negative rate — the misses — since "the system didn't catch it" is only a defense if the system was demonstrably maintained?
Notice that none of these are AI questions. They're controls questions — the same questions a fraud examiner or forensic accountant asks about any process: who was accountable, was there a control, and can you prove it operated? That an algorithm is involved doesn't change the discipline. It just raises the stakes, because now the process documents itself.
Who should be asking these questions
Not the company that sold you the safety system. Their job is to make it detect well, and they're incentivized to report that it's working. The party who evaluates whether your deployment is defensible has to be independent of the party who built it — for the same reason a business doesn't audit its own books. Independence isn't a formality here. In a courtroom, self-assessment carries almost no weight; an independent review carries a great deal.
For a non-subscriber, this is not an abstract governance nicety. It's the difference between a safety investment that protects you and one that, on the worst day, testifies against you.
Is your safety AI proof — or exposure?
The AI Governance Readiness Assessment is a focused, independent review built for exactly this question. No downtime, no code access, no committee. I look at what your system flags, whether those flags get a documented human response, and whether the record it's building would help you or hurt you in a dispute — then hand you a short findings report rating your exposure and the smallest set of fixes that make it defensible.
Independent by design. I don't sell safety systems, so I have no stake in telling you yours is fine. As a CPA (Ret.) and Certified Fraud Examiner, the questions I ask are the ones a plaintiff's attorney will — before they do.
Request an assessment
This article is informational and describes governance and controls practices. It is not legal advice and does not create an attorney-client relationship. For questions about your obligations as a Texas non-subscriber, consult qualified counsel.