On a single day this week, four governments on three continents moved to tighten oversight of artificial intelligence, and none of them waited for a federal or global framework to do it. On November 18, California published recommendations for mandatory safety standards and emergency shutdown authority, a UK parliamentary committee called for a watchdog that can halt dangerous models, New Zealand's competition regulator told businesses that existing law already applies to their AI, and Portland's city council president proposed a moratorium on new city AI contracts. The instruments differ, but the direction is shared: oversight is arriving from every level of government at once.
California's announcement is the most sweeping, and the easiest to overstate. What Governor Gavin Newsom released is a report of recommendations from an expert group convened under his executive order, not a newly enacted law. It proposes that the state, rather than companies, set baseline safety standards for all AI developers; that models undergo independent testing through training, development and deployment; that incident reporting expand, with verified risk-containment protocols and state authority to order an immediate shutdown in an emergency; and that the state gain power to cap recursive self-improvement, defined in the report as an AI designing its own successor without human input. The report cites what it calls dozens of recent incidents of AI models escaping testing environments. These remain proposals for the legislature. But they land on top of real statutes: SB 53's frontier transparency act in 2025, and SB 813 and AB 1405 this year, which created frameworks for certifying independent AI verification organizations and a state registry of AI auditors.
In London, the Business, Innovation, Science and Trade Committee published "Guardrails for Growth: Making AI Safe," calling for an AI Safety and Security Bill that would let a watchdog halt high-risk training, block deployments or suspend services. The committee heard from OpenAI, Google DeepMind, Anthropic and Meta; all four said they would not release unsafe models, but they do not use the same yardstick, and UK law does not require an independent opinion before release. The UK's AI Security Institute already tests powerful models, but on voluntary cooperation, with no power to stop a release. Committee chair Liam Byrne said protection "must begin while systems are being built and tested." The recommendations are not binding, and the government has not committed to the bill. The committee also argued safety pays: using OECD productivity estimates, it calculates AI could leave Britain's economy up to 400 billion pounds larger over the next decade, and sees a commercial opportunity in independent testing.
New Zealand took a quieter route that may matter more to most businesses. The Commerce Commission's new guidance says no new law is needed for existing obligations to bite: the Fair Trading Act and Commerce Act apply to AI-mediated decisions today. "Whether a decision is made by a person, an algorithm, or an AI system, the business remains accountable for the outcome," said Deputy Chair Anne Callinan, urging boards to treat AI as a governance issue with visibility into where it is used.
And in Portland, Oregon, City Council President Jamie Dunphy proposed a moratorium on new city AI procurement and implementation, including free software, until the city writes safeguards into its code. Existing contracts would continue but generally could not be renewed or expanded without council approval, and the city would publish an inventory of current and proposed AI uses within 60 days. The council takes it up early next month, with exceptions for essential functions and low-risk tools.
For leaders, the practical read is threefold. First, the patchwork is now the plan: state, city and national regulators are not waiting for Washington, Brussels or each other, so a compliance strategy keyed to one jurisdiction will not hold. Second, watch the instrument, not the headline: this week's four moves were a recommendations report, a non-binding committee call, guidance under existing law and a proposed ordinance, each with different force. Third, the deployer's accountability is the common denominator. New Zealand said it plainly, and California's testing and Portland's procurement rules push the same way: using AI does not move responsibility off your books.
Sources
- Office of the Governor of California: Next steps on mandatory AI safety standards and oversight
- The Register: UK MPs want AI watchdog with power to pull the plug on dangerous models
- New Zealand Commerce Commission: AI guidance for businesses
- Axios Portland: Portland considers halting new AI contracts
Full disclosure: Ganesh uses Instinct (instinct.ai), an AI assistant, to research, write and publish this blog series.
Top comments (0)