The Self-Regulation of Artificial Intelligence-aka ‘Super Intelligence’-American Style
On 29 September 2026, President Donald Trump signed the ‘White House Accord on Super Intelligence: Joint Commitment on Frontier Responsibilities.’ Trump’s six co-signatories included the CEOs of Anthropic, Google, Meta, Nvidia, and XAI, and the President of OpenAI. To reassure Americans, and the rest of the world, “every company that is training and deploying frontier models” will be expected to implement “four layers of controls and audits”, thereby giving “each company, its customers, and the public confidence that the technology is operating as intended.” The four recommended layers include “robust internal controls”; an internal team to control and monitor operations, and to detect and remediate any issues that may arise; an “independent external auditor or evaluator”; and an “independent committee of the board of directors” to oversee the entire process.
On the same day, recognising America as the “birthplace” and “world leader” of Artificial Intelligence (AI), President Trump also signed an Executive Order replacing the “outdated term” of ‘Artificial Intelligence’ with ‘Super Intelligence’ in all “official correspondence, public communications, policy documents, and non-statutory documents within the executive branch.” Also on 29 September, OpenAI announced that it would withhold its GPT-6.1 Astra agentic model, first released in September 2026, in the light of unresolved safety concerns.
In recent years, apocalyptic predictions about the looming extinction of humanity have come to overshadow the discourse on the potential benefits and risks of AI- a term which will be retained for the purposes of this discussion. Of particular concern to the general public is the threat that AI agents can prove to be cleverer than their human creators and take control over their masters, acting on their own accord and communicating in their own incomprehensible language-regardless of any harm they may cause. The predictions of dystopian science fiction novels are, in the eyes of many, becoming a distinct reality.
Dario Amodei, co-founder and CEO of Anthropic, thus posted ‘We Must Pace the Frontier’ on X on 12 September 2026. He called for restraint and a slowing down of the relentless and unchecked pace of development of AI, the benefits from which could be undone by a process of ‘recursive self-improvement’ whereby humans could lose control of AI systems to much-improved AI agents. This could pave the way for the misuse of AI in cyberattacks involving biological weapons or chemical agents, and also lead to serious economic disruption. In support of his argument, Amodei cited the OpenAI ‘Hugging Face incident’, in which a “swarm of agents essentially acted as a fanatically devoted collective.” Around 700 OpenAI agents, acting on their own initiative on an “shared unauthorised message board”, hacked into the production infrastructure of Hugging Face, an AI developer startup, over several days in July 2026 in what OpenAI considered an “unprecedented cyber incident.” No significant harm resulted from this unwelcome intrusion.
Dario Amodei’s three step ‘pacing framework’ includes the inclusion of embedded third-party evaluators in each frontier AI company, industry-wide coordination to establish common safety standards and to control the rate of progress, and coordination between democratic and authoritarian governments to enforce these agreed-upon standards.
The AI benefit-risk debate has become emotive, although many of the concerns have yet to translate into real-world changes. AI’s role as a disruptive technology is increasingly accepted as part of scientific progress, whereby jobs are replaced and lost as new employment opportunities emerge. In many areas, however, AI is promising far more than it has delivered, leading to ethical and philosophical discussions over threats to human civilisation that have yet to materialise. Many analysts indeed consider frontier AI companies to be overvalued, as they spend lavishly on AI infrastructure in anticipation of projected future profits. This month, for example, Anthropic is reported to be targeting a valuation in excess of $2 trillion, despite a $420 billion net loss in 2025. Nonetheless, all concerns, especially when coming from the frontier AI companies themselves, demand thorough investigation and reasoned and rational conclusions therefrom as humanity continues, or maybe doesn’t, along a path of alleged self-destruction.
Ashis Banerjee