President Trump has proposed a voluntary self-regulation accord with technology executives that would let companies set their own safety standards for artificial intelligence, according to the facts of the proposal. The plan, which would rely on industry to police itself, has drawn immediate concerns about accountability and transparency in AI development.
What the accord would do
The proposal asks tech leaders to sign onto a shared set of principles for AI safety, with companies responsible for deciding how those principles get applied. It doesn't create a new federal agency or grant regulators new powers. Instead, it leans on the companies building the most advanced AI systems to monitor their own work and report on it as they see fit.
That approach is a departure from the more prescriptive rules that some lawmakers and safety advocates have pushed for in recent years. It also puts the White House in the position of trusting the same firms that stand to profit from moving fast on AI to slow down when risks emerge.
Why self-regulation worries critics
The core objection is simple: a company that grades its own homework has little incentive to fail itself. Without outside audits or enforceable penalties, safety measures are likely to vary widely from one firm to the next. What one company treats as a red line — say, releasing a model with known flaws — another might wave through to stay competitive.
Accountability is the other sticking point. If an AI system causes harm, a voluntary accord doesn't clearly say who answers for it. The company? The executive who signed? The White House that brokered the deal? The proposal as described doesn't resolve that, and that ambiguity is exactly what critics say will make the accord hard to enforce.
Transparency is a related problem. Self-reported safety data can be selective. Companies can highlight their best practices and stay quiet about incidents that never make it into a public report. Without a common disclosure standard, comparing one firm's safety record to another's becomes nearly impossible.
The competitive pressure problem
Even executives who genuinely want stronger safeguards face a collective-action problem. If one company spends months on safety testing while a rival ships a similar model sooner, the cautious firm loses ground. A voluntary accord doesn't change that math unless every major player signs on and sticks to it — and there's no mechanism in the proposal to force either.
That's not a hypothetical. The AI industry is moving quickly, and the firms at the table are direct competitors. A handshake agreement among rivals is only as strong as the least committed participant.
What happens next
The proposal now goes to the tech executives for their response. Whether they sign on, push for changes, or decline will shape what the accord actually looks like — and whether it has any teeth. Until those signatures are on paper, the safety standards it describes remain a draft, not a rule.
