I keep thinking about the laminated sign at the public pool where I grew up: no running, no glass, no diving in the shallow end. Nobody confused it for political theory. It was just the social contract required to let a lot of excited people share a risky rectangle of water. AI now has a pool sign problem, except the strongest swimmers are offering to write the sign, install the fence, and define what counts as safe behavior.

The Patch Before the Rulebook

MIT Technology Review reported that on July 21, 2023, seven leading AI companies, Amazon, Anthropic, Google, Inflection, Meta, Microsoft, and OpenAI, made eight voluntary commitments with the White House on developing AI in a safe and trustworthy way. Those commitments included improving testing and transparency around AI systems, plus sharing information on potential harms and risks, according to the publication. One year later, MIT Technology Review found welcome progress on practices such as red teaming and watermarks, but said there was no meaningful transparency or accountability.

That is the tension hiding inside the phrase voluntary commitment. It can be genuinely useful when lawmakers are moving slowly and product teams are shipping quickly. MIT Technology Review also noted that eight more companies later signed the White House commitments, and that an executive order expanded on them with a requirement to share safety test results for new AI models with the US government if the tests show national security risk. A patch can fix an urgent bug, but it also tells you who has commit access.

When Coordination Sounds Like Safety

The optimistic case for self-regulation was captured by BBC Sounds, which reported Sam Altman saying AI companies like his were capable of essentially regulating themselves. Altman said, "We will get it right, I'm very confident in our company's and industry's ability to do this safely," adding that companies would "keep alignment and safety way ahead of capabilities" and, if they could not, would "slow down or stop." That is a powerful promise because it frames coordination as restraint rather than control.

There is a serious version of that argument. MLex reported in November 2024 that US AI companies said they were working to fulfill voluntary safety commitments made to the White House, including providing information to government and the public about their technologies and risks. For builders, that kind of coordination can create shared expectations around testing, disclosure, and incident response. The problem begins when a safety norm becomes a market gate, and the people who helped define the gate are also the ones best equipped to pass through it.

The Market Hidden Inside the Rule

MLex reported in September 2026 that leading AI companies were pushing for industry-led safety standards for frontier AI in the EU. That phrase, industry-led, is doing a lot of work. It can mean the people closest to the systems help regulators understand fast-moving technical risk, or it can mean the largest firms turn their own operating model into the default compliance model.

WIRED reported that Anthropic supported the first wave of frontier AI safety laws in the United States, including transparency requirements in California and New York that much of Silicon Valley opposed on the grounds that they could stifle the AI boom. WIRED also reported that Anthropic is now pushing states toward tougher regulations, while saying that transparency and self reporting are no longer sufficient for the most powerful AI systems. That makes Anthropic an unusually clear example of the modern AI governance puzzle: a company can be sincere about safety and still benefit from rules that smaller rivals may struggle to satisfy. Governance is not just ethics in a suit; it is also go-to-market strategy with public-interest vocabulary.

What Builders Should Watch Next

The uncomfortable question is not whether AI leaders believe what they are saying. Some clearly do, and MIT Technology Review’s reporting suggests voluntary commitments have nudged parts of the industry toward better practices. The sharper question is whether coordination creates independent accountability or mainly produces a club of firms large enough to coordinate. If the standard requires expensive testing, specialized legal teams, and constant regulator access, safety can become a moat even when nobody calls it one.

For product leaders, the practical move is to separate safety outcomes from institutional convenience. Ask whether proposed rules publish enough information for outsiders to evaluate compliance, whether audits can be performed by independent parties, and whether smaller developers have a credible path to meet the same bar. For investors, watch whether regulation raises trust in the category or simply narrows the field to companies already sitting on compute, distribution, and policy relationships. The next frontier AI rulebook may genuinely reduce risk, but it will also decide who gets to keep building, so when the biggest labs ask to coordinate, are we watching safety infrastructure take shape, or the future market drawing its own borders?

Sources