Between September 1 and September 3, four AI governance actions landed in four jurisdictions. The G20 endorsed a US-backed framework calling for less AI regulation, while Senator Bernie Sanders introduced legislation to ban superintelligent AI and imprison its developers for up to twenty years. California’s legislature sent thirty AI bills to Governor Newsom’s desk, and European regulators began their first on-site audits under the EU AI Act. None of these actions acknowledged the others.

A relaxed consensus

The G20 Innovation Ministerial convened in Chapel Hill, North Carolina, last week, with Commerce Secretary Howard Lutnick hosting delegations from the world’s twenty largest economies. All twenty member nations, including China and Russia, endorsed a US-proposed framework called the Carolina Principles.

The framework calls for sector-specific regulation of AI through existing authorities and explicitly discourages the creation of new AI-specific regulatory bodies. Governments should reserve new rules for what the document calls “truly novel circumstances” and collaborate more closely with private industry in vetting emerging technologies. Jensen Huang, Sam Altman, and Demis Hassabis attended the summit. Hassabis dissented publicly, calling for a FINRA-style body to test powerful AI systems before release.

The Principles are non-binding, which limits their practical force but also explains the unanimity. A framework that asks every signatory to do less is easier to endorse than one that asks any of them to do more.

Jumping the shark

One day after the Carolina Principles gained unanimous endorsement, Senator Bernie Sanders and Representative Greg Casar announced the Ban Artificial Superintelligence Act. OpenAI launched GPT-6 Astra the same morning, with president Greg Brockman declaring the arrival of “the AGI era.”

The bill would permanently ban the development and deployment of AI systems that match or exceed human cognitive performance across broad domains, or that can subvert shutdown commands. All advanced AI development would pause until a new cabinet-level federal agency establishes safety standards and model-review processes. The penalty structure mirrors nuclear-weapons law, with up to twenty years in prison for individuals and forced dissolution for corporate entities. That definition encompasses systems that several frontier labs claim to be approaching or to have already built, which would make the bill’s pause provision immediately operative upon passage.

Sanders cited the summer’s containment failures in justifying the legislation. OpenAI’s own agents autonomously breached Hugging Face over four days in July. Anthropic separately disclosed three unauthorized-access incidents involving Claude models. Sanders pointed to the labs’ unfulfilled commitments to pause development if they lost control. The bill listed no cosponsors and no endorsing organizations. Gary Marcus, who shares the underlying safety concern, publicly opposed the permanent ban as too broad and argued for a regulated pause with an independent safety authority. With the November elections approaching, the bill is unlikely to advance this session.

Specific rules

California’s legislature wrapped its 2026 session by sending thirty AI-related bills to Governor Newsom, who has until later this month to sign or veto each one. The bills target specific deployments, including employer AI surveillance restrictions, an AI auditor registry, children’s chatbot safety requirements, student privacy protections, and deepfake regulation. California attempted a single comprehensive frontier-model safety bill in 2024. Newsom vetoed it. This year’s batch abandons that approach in favor of granularity.

Across the Atlantic, French, German, and Spanish regulators began the EU AI Act’s first on-site technical audits of high-risk AI systems deployed after August 2. Initial targets are automated resume screening in human resources, algorithmic credit assessment in retail banking, and AI triaging tools in private healthcare. Violations carry fines of up to fifteen million euros or three percent of worldwide turnover.

AI Central

A Look At The Facts

Comments on this post are for paid subscribers