The consensus feels reassuring: major AI companies are getting serious about safety. They're running security tests. They're hiring ethics teams. They're publishing responsible use policies. Good news, right?
Not quite. The better question is what this performative safety focus breaks in how we actually hold corporations accountable.
Consider what's happened in recent months. We've seen companies conduct internal security audits, only to discover their own models can breach customer systems. We've watched executives publicly call for regulatory slowdowns while their investors place massive bets on acceleration. We've observed startups market themselves on founders willing to work grueling schedules, framed as commitment rather than red flag. The pattern suggests something important: safety language has become a corporate permission structure rather than a constraint.
Here's the trap. When a company announces a safety initiative, the narrative shifts from "we need external oversight" to "we're handling this internally." It's not that safety work is inherently bad. It's that the visibility of internal efforts obscures the actual absence of external accountability mechanisms. A security test conducted by the company building the product is useful data. But treating it as evidence of solved problems is exactly backwards.
The real stakes emerge when we ask: what breaks if we accept corporate safety efforts as sufficient?
First, the investor time horizon breaks. A company genuinely constrained by safety concerns would grow slower. It would pause before deployment. It would prioritize caution over market position. Investors would accept lower returns. But we don't see that trade-off happening. Instead, we see safety rhetoric coexist with "move fast" cultures, with founders openly discussing extreme work demands, with funding rounds that price in aggressive expansion. This tells us which value actually wins when they conflict.
Second, regulatory clarity breaks. When companies self-govern, regulators face an impossible problem. Do you trust internal audits? Do you impose external ones? By the time you answer, the company has already shipped to millions of users. The regulatory vacuum doesn't stay empty. It gets filled by whoever moves fastest, which is rarely the most cautious actor.
Third, and most subtly, public trust in institutions breaks. Safety theater works only if people don't look too closely. But they will. Eventually, they'll notice that every serious problem discovered came from outside scrutiny, not internal processes. They'll see the gap between stated values and incentive structures. And they'll lose faith not just in tech companies, but in the very idea of corporate responsibility.
The companies doing this aren't necessarily acting in bad faith. Many genuinely believe their internal processes work. But good intentions don't create accountability. Visibility does. Consequences do. Independent verification does.
What should actually concern us isn't whether AI companies care about safety. It's whether the current system gives them strong reasons to prioritize it when safety conflicts with growth, speed, or profit. Right now, it doesn't.
The consensus says we can trust companies to police themselves if they hire the right teams and publish the right policies. The better question is simpler: why would we expect that to work when it hasn't worked for any other industry?
The path forward isn't rejecting corporate safety efforts. It's recognizing them for what they are: a necessary starting point, not the finish line. Real accountability requires something companies can't provide internally: external leverage, clear standards, meaningful consequences for failures, and skeptics with power asking hard questions.
Until that changes, safety language is just another corporate tool. And the real risks are hiding in plain sight.