OpenAI pushes Congress for federal AI safety rules, endorses four California bills

OpenAI published a policy statement arguing that the window to put safeguards around frontier AI in place is closing, calling on Congress to pass mandatory federal AI safety rules and formally endorsing four pending California AI bills.

What's new

On the federal side, OpenAI is pressing Congress to move past voluntary commitments. In its own words: "The United States needs mandatory, capability-based national regulation that can evolve as the technology does." The company wants requirements pegged to what a model can actually do, rather than fixed rules that age out as capabilities advance.

At the state level, OpenAI formally endorsed four California bills:

• SB 813 — establishes independent safety assessment processes for frontier AI systems.
• AB 1405 — creates AI auditor standards and accountability requirements.
• SB 1119 — requires protections for minors using chatbots, including age assurance, risk assessments, independent audits, and parental controls.
• AB 1864 — strengthens safeguards against AI-enabled biological threats.

On industry practice, OpenAI says it is committing to build shared monitoring frameworks for model "misalignment" together with other frontier labs, and to support "compatible international approaches" to measuring capability and managing risk across jurisdictions. Notably, the company says it will back rules that are "not exactly what we would have designed" as long as they meaningfully raise the safety bar — a concession that it expects to lose some fights over the specifics.

Context

The post lands in a week when OpenAI's own chief scientist, Jakub Pachocki, said publicly that no lab has yet solved alignment for safe scaling, and just after Anthropic warned about the risk of imminent recursive self-improvement in frontier systems. OpenAI has also previously reversed course on state regulation, calling on California to strengthen SB 53 rather than oppose it, after early criticism that the industry was resisting oversight. This new post extends that posture: rather than reacting bill-by-bill, OpenAI is laying out a standing position — endorse capability-based state rules now, and ask for a federal floor before a patchwork of state laws hardens into the default.

Why it matters

A public, named endorsement of four specific bills is a stronger commitment than the general safety rhetoric OpenAI and its peers have offered in the past — it creates a paper trail lawmakers and reporters can hold the company to if it lobbies against similar measures elsewhere. It also reflects a broader shift among frontier labs from defending against regulation to shaping it before Congress or European regulators do it without their input. The emphasis on "capability-based" rules in particular signals where OpenAI wants the eventual federal standard to land: thresholds tied to what a model can do (cyber, bio, autonomous replication) rather than blanket rules on all AI, which would be far more favorable to a company shipping frontier systems on a rapid cadence than a static compliance regime designed around today's models.