No AI Safety Rule Binds Your Vendor Yet

Three Mechanisms, One Week, No Binding Force

In the week of September 9 to 15, 2026, three distinct mechanisms for policing frontier AI safety surfaced in the United States, and none of them carries the force of law. A bipartisan bill picked up a public endorsement from an OpenAI executive, a billionaire pitched a voluntary testing pact at a Los Angeles conference, and a rival lab's own chief executive said in public that his company does not even need the antitrust waiver its industry peer had requested.

None of these three is a rule yet. The FRONTIER Act is a bill awaiting a vote, Musk's peer-review pitch has zero companies signed on including his own xAI, and the antitrust waiver Anthropic requested has not been granted. For a European or British company weighing a US AI vendor's safety pledge this month, that is the fact worth sitting with: the pledge is a voluntary position inside a live three-way argument about what the actual mechanism should even be.

The FRONTIER Act Waits on a September Markup

Representatives Jay Obernolte and Lori Trahan introduced the FRONTIER Act, formally H.R. 9925, in the US House on July 23, 2026, as a bipartisan bill aimed at the handful of companies building the most powerful frontier AI models. The bill sets a risk-based, tiered framework requiring model cards, risk-management frameworks, independent audits carried out by what the bill calls Independent Verification Organizations embedded inside the largest labs, incident reporting for serious safety events and ongoing assessment of catastrophic risk, and it is meant to give developers one federal standard instead of a state-by-state patchwork.

"The FRONTIER Act focuses oversight on the largest developers and most advanced models, requiring transparency, independent evaluation, and timely reporting of serious safety incidents," Obernolte said. Trahan called it a way to deliver "commonsense transparency and independent oversight for the largest AI developers while giving them a single, clear national standard to build on." On September 15, 2026, OpenAI's head of global affairs Chris Lehane said publicly that OpenAI supports the bill's independent-evaluator provision, and lawmakers had pointed to a September markup, but the bill has not passed either chamber and today it enforces nothing.

Musk Wants Rivals to Test Each Other's Models

Elon Musk proposed a voluntary peer-review system for frontier AI models at the All-In Summit, a technology conference in Los Angeles run by the All-In podcast hosts, which took place September 13 to 15, 2026. Appearing in a session with SpaceX president Gwynne Shotwell, Musk suggested that leading AI companies test each other's models using each lab's own existing safety-test harness rather than each lab grading its own homework, restating a framing he had given in an earlier interview on July 23, 2026: leading AI companies should at least meet every few weeks to discuss safety and security issues, and rivals should get a week or two of early access to a competitor's new model before release so that "competitors can keep each other honest."

Musk said government should step in only "as a last resort," if a company ignored a safety flag raised by a rival, and added that any such system has to be something China is willing to accept to mean anything, noting that oversight is easy to add and hard to remove. No company, including Musk's own xAI, has agreed to the scheme, there is no mechanism for running it, and no timeline exists for turning the pitch into practice.

The Antitrust Waiver Nobody Has Granted

Anthropic chief executive Dario Amodei asked the US government for a narrow antitrust waiver so that OpenAI, Anthropic and Google DeepMind could legally coordinate on safety standards without running into competition law, building on informal three-lab talks about a safety-standards body that have continued since around July 2026. OpenAI chief executive Sam Altman publicly rejected the idea that the waiver is even necessary, saying, per reporting, that "OpenAI welcomes consistent federal requirements, yet does not need the waiver or new legislation to start."

No waiver has been granted, and the three-lab talks remain informal discussions rather than a signed agreement of any kind. Altman's position means OpenAI intends to keep building its own voluntary safeguards regardless of whether Washington ever grants the waiver Amodei asked for, which leaves the coordination Anthropic wanted sitting exactly where it started.

Three Mechanisms, Side by Side

Laid next to each other, the three approaches share a subject and nothing else: a pending bill, a voluntary pitch with no signatories and a legal request that has gone nowhere.

For an EU or UK operator buying from any of these vendors, the contrast that matters is not among the three US proposals but against the one framework that already binds them: the EU AI Act's general-purpose AI obligations are in force now, while every US mechanism above is still an argument about what the rule should even be. A vendor's safety pledge issued this month is a voluntary position taken inside that argument, and treating it as a settled regulatory floor gets the contract terms wrong.

Servola Journal

We do this for everyone trying to keep up with what technology is doing to our lives. The people who build it, and the people it happens to. The Servola Journal exists so that what we learn belongs to all of them.

Nobody pays us for this. No ads, no paywall, free to everyone. We just believe that understanding what's happening to all of us shouldn't depend on who can afford to pay for it.

If it gave you something today, tell us to keep going. Follow us, leave a like, or write a positive comment. We read every one, and they are what keeps us going.

Read next: US Bill Would Jail AI Developers for 20 Years  |  A Security Badge Told The Pentagon Nothing About What Grok Would Generate