samlevine277.github.io

AI Safety Before Release: Why Customers Deserve Choice

January 20, 2026

Artificial intelligence is being released faster than ever, and companies often claim they are improving safety through user feedback. But I believe there is a dangerous assumption behind that idea: that it’s okay to use real customers as unpaid testers. When AI systems are released without proper testing, people can be exposed to serious harms—misinformation, bias, privacy violations, and emotional manipulation. Safety should not be something we “learn” by watching users get harmed.

In Dr. Plate’s post “Two Theories of Safety,” he explains that OpenAI and Anthropic disagree not on whether safety matters, but on how safety is discovered. One founder believes safety emerges through deployment, while the other believes safety must be proven before release. I think this disagreement matters because it reveals a deeper question: should the public be forced to test dangerous technology?

“Both founders believe AI should be safe. Neither is lying about this. But they hold different theories about how safety is achieved — and those theories trace back to their backgrounds, their training, what they learned before they ever touched artificial intelligence.”
— Dr. Plate, “Two Theories of Safety”

I agree with Dr. Plate that both founders care about safety. But caring is not enough. The question is whether it is ethical to release a product before its risks are fully understood. My position is simple: AI should be tested and proven safe before it is released to the public. If a company wants to launch an AI tool early, they must clearly tell customers that it is still in testing and allow them to choose whether to use it.

A company that releases an untested AI system is essentially saying, “We don’t know what could go wrong, but you can find out for us.” This is unfair because users are not paid testers, and many don’t understand the risks. If an AI system is flawed, the harm can be immediate and irreversible. A customer might rely on AI for medical advice, mental health support, or financial decisions. If the AI fails, the consequences are real and dangerous.

If this happened to me as a paying customer, I would be extremely upset. I would feel like the company used me as a test subject without consent. I would lose trust in that company forever. Knowing that a product was released without testing, and that the company still charged users for it, would feel like a betrayal. Trust is not something companies can regain easily once it is broken.

Some argue that AI must be released to discover flaws that internal testing misses. But the reality is that real users shouldn’t be the ones discovering those flaws. That’s why high-risk industries don’t operate this way. Cars, airplanes, and medical devices must be tested before they hit the market. Why should AI be treated differently? The potential harm is just as serious.

Evidence That Safety Must Be Proven First

Researchers in AI safety have proposed formal frameworks that require evidence of safety before deployment. One such approach is called “affirmative safety,” which argues that high-risk AI systems should provide proof that their risks remain below acceptable levels before being released. This model emphasizes evidence and accountability, rather than relying on users to discover problems after deployment.

“Entities developing or deploying high-risk AI systems should be required to present evidence of affirmative safety: a proactive case that their activities keep risks below acceptable thresholds.”
— Akash R. Wasil et al., Affirmative Safety: An Approach to Risk Management for High-Risk AI (2024)

This quote supports my argument that AI companies should prove safety first. If a system cannot show that it meets safety standards, it should not be released to the public. Releasing it anyway is not innovation—it’s risk shifting. The company is shifting the burden of testing onto customers who did not consent to being part of a dangerous experiment.

There is also a trust issue. When companies release AI before testing, they break trust with users. People should be able to use technology without fear of being harmed or manipulated. If a company is transparent about its safety status, users can choose a safer alternative or wait for a better version. This transparency protects consumer rights and forces companies to be responsible.

Why Customers Must Have Choice

If an AI product is still in testing, customers should be told clearly and given the option to opt out. The problem is that many companies hide this information or bury it in legal terms. That is not consent. Real consent means customers understand the risk and choose freely.

I believe the public should not be treated as a testing ground. If AI companies want to use real users for testing, they must pay them and inform them clearly. Otherwise, the public is being exploited. And that is not safe, ethical, or fair.

In addition, testing before release does not mean companies cannot innovate quickly. They can still iterate in controlled environments. They can use staged rollouts, private beta testing, and independent safety audits. These methods allow innovation without exposing the public to unnecessary harm.

Conclusion

AI has the potential to change the world, but that power comes with responsibility. The debate between “deploy first” and “prove safety first” is not just a technical argument; it is an ethical one. I believe companies should not release AI systems without proven safety. If they choose to release early, they must be transparent and give customers a real choice. Otherwise, they are using the public as unpaid testers, which is unfair and dangerous.

Safety is not just a buzzword — it is a requirement. The public deserves technology that is both powerful and safe. If AI companies want to be trusted, they must prove that their products are safe before asking users to risk their lives, privacy, and well-being.