News

Stephen Pitt-Walker, JD, FGIA, CEO & Founder, Optima Board Services Group.

gettyIn July 2023, the United Nations Security Council held its first session devoted to artificial intelligence (AI). Secretary-General António Guterres warned that AI could accelerate human development while threatening democratic institutions and social stability if left ungoverned. Now, three years later, leaders in tech are sounding the alarm, too. While the world’s governing bodies work to agree upon a set of universal AI rules, corporate boards should also be discussing what their organizations can do to ensure responsible use.

In some places, AI oversight already falls within directors’ existing duties. For example, in Australia, section 180 of the Corporations Act requires directors to exercise care and diligence, and in Delaware, Caremark oversight duties apply. ​Here’s what leaders should know about the ongoing AI pacing debate, as well as some steps boards can take today:​

Why Tech Leaders Are Making A Case For Slowing Down​On Sept. 12, 2026, Anthropic chief executive Dario Amodei published “We Must Pace the Frontier,” asking the industry to slow the rate at which frontier model capabilities improve. Amodei pointed to two developments as the reason for more restraint. First, AI systems had begun materially accelerating the development of their successors. Second, during a July test, OpenAI agents attacked Hugging Face systems in a move unrelated to its task. Amodei’s suggested approach is more measured than the “slow down” label suggests. Pacing means progress at a governed rate. Anthropic committed to embedded third-party evaluators with employee-level access, modeled on banking supervision. ​

OpenAI’s Sam Altman and xAI owner Elon Musk backed him almost immediately. It was striking that the people building the technology were asking, in public, for restraint. ​Critics likened the tech leaders’ actions to regulatory capture, and on Sept. 18, consumers filed an antitrust class action alleging an unlawful agreement to slow development. The legal lesson is clear: Coordination among competitors on safety needs public authority to be durable.

The Fractured Response From Governing Bodies​Within days of Amodei’s request, the major powers responded, exposing the fault lines in global AI governance. While U.S. President Trump wants to keep the strategic edge the U.S. has in AI, declaring that “whoever wins AI wins,” he does support some safety measures. China’s Foreign Ministry dismissed the warnings as fearmongering, and China’s state media cast the proposals on chip exports and model copying as a Cold War playbook, even as Chinese President Xi has named loss of control as a risk requiring international rules. A September meeting at the White House between Trump and Xi led to a new US-China “Super Intelligence Dialogue,” intended to serve as a bilateral channel for AI incidents. The summit extended the trade truce but so far has produced no agreement to slow development or constrain AI risk.​

Meanwhile in Europe and Australia, leaders backed restraint but lacked the leverage to impose it. The European Commission President Ursula von der Leyen endorsed pacing in her State of the Union address. Yet the EU already deferred its AI Act’s high-risk obligations to late 2027, and most frontier developers are based outside Europe. Australian Prime Minister Anthony Albanese joined around 20 other countries and the European Commission in a declaration insisting that AI remain under human control. His disclosure that an OpenAI agent had accessed non-public files on a Medicare portal, with notification arriving months later, gave that argument moral force, and he carried it to the UN General Assembly.​

Borderless Technology, Territorial lawA tripolar order is now in plain view. The EU governs to protect rights, the United States to secure innovation and primacy, and China to preserve political stability and maintain competitiveness. Middle powers are building coalitions in the interstices. Yet AI is borderless and law is territorial. An agent tested in one jurisdiction can breach systems in another, as the Medicare incident and others reportedly show.

Precedent indicates the gap can narrow. The Basel Committee coordinated banking supervision across borders, and the Council of Europe’s 2024 AI convention shows that binding agreement on principles is attainable. Arms control suggests the sequence: Narrow prohibitions first, then common pre-release testing, then limits on recursive self-improvement. The goal is interoperability over uniformity. That work will take years, and boards cannot wait for it.

What Boards Should Request NowThe pacing debate sharpens three questions for boards:

1. Can management evidence independent verification of the AI systems the organization builds or relies on?

2. Do incident protocols guarantee prompt, direct notification to regulators and affected parties?

3. Where the organization joins industry safety collaboration, has its competition law exposure been assessed?

Put robust answers on the record. Minute the evidence, give each material AI system a named executive owner and set a reporting cadence so the board sees incidents and near misses, not just assurances. Weak answers call for proportionate action. If management cannot evidence independent verification, commission it, starting with customer-facing and regulated systems, and make continued deployment conditional on the results.

If incident protocols are vague, set a notification standard measured in days, test it with a tabletop exercise and write the same obligation into AI vendor contracts. Much of an organization’s exposure subsists in systems it buys. If competition exposure has not been assessed, take advice before joining industry safety initiatives and document the safety rationale. Then, close the capability gap at the top. A board that cannot interrogate AI cannot oversee it, so invest in director education and give AI risk a standing place on the risk committee’s agenda.

Lastly, boards can push for better global policies. Responsible AI requires more than the goodwill of developers or the priorities of any single capital. It requires law. The role of law is to civilize technological progress, and the stakes with AI are civilizational. If AI is to serve humanity, humanity must first decide how to govern this form of intelligence. For most organizations, that decision begins at the board table.​

Forbes Business Council is the foremost growth and networking organization for business owners and leaders. Do I qualify?