Limits on delegating decision to the machine: Sam Altman makes human control a divider between OpenAI and its competitors

OpenAI CEO Sam Altman opened his participation at the Abu Dhabi Adibak conference with a decisive statement on the issue of delegating decision-making to intelligent systems, confirming in his interview with The National editor Mona Al-Arabi that there is an indisputable principle:Control of the future must remain in the hands of humans, not artificial intelligenceThis stance did not arise in a purely theoretical context; Altman presented it as a decisive line that distinguishes his company from other competitors, stressing the rejection of any full handover of leadership to algorithms, and affirming that technology is designed to serve humans, not the reverse.

These remarks come at a moment Altman describes as a pivotal turning point in the trajectory of models, noting that artificial intelligence has leapt this year from lagging behind the world’s brightest scientists to surpassing them in most tests, alongside its rising ability to perform large portions of office work. While acknowledging society’s capacity to adapt to these changes, the OpenAI chief warned that the greatest burden now falls on the tech sector to provide practical safeguards that prevent serious incidents related to system safety and alignment with human values, questioning how companies can convince the world of the effectiveness of these safeguards as model capabilities multiply.

During the session Altman revealed an experiment in which OpenAI used ten thousand intelligent agents working in parallel under the anticipated GPT model to solve one of the “Millennium Prize” mathematical problems over four consecutive days of horizontal collaboration without any hierarchical structure, an event that has not yet been independently verified. The demonstration also shows the enormous scale of resource consumption; while the company’s most resource-intensive internal user six years ago burned one hundred thousand tokens per month, roughly the current global average per person, the largest user today consumes nearly five hundred billion tokens per month, explaining Altman’s terse reply when asked about the biggest challenge facing him in one word: energy.

Altman believes that the cost of inference per query has fallen dramatically, but the explosion in demand has made electrical power the decisive factor in determining the global scale of available artificial intelligence. The company’s chief hopes that these systems will eventually contribute a net addition to energy production, although he acknowledges thatreducing computing costs faces the reality that physical atoms remain subject to the laws of physics. Within this framework OpenAI’s recent agreements in Abu Dhabi gain practical weight, especially the collaboration with ADNOC and the G42 technology group to harness the latest advanced models for tackling the most complex problems in the energy sector, as well as education, investment and government services, benefiting from a high adoption rate that his estimates indicate that about eighty percent of Emirati youth use the ChatGPT tool.

This approach places the digital transformation leaders and engineering teams in the Gulf, Egypt and the Levant before a clear operational equation. Linking the latest models to heavy-energy sectors and institutional infrastructure means that the next wave of investment in the region will not stop at ready-made conversational interfaces, but will head straight toward swarms of software agents that manage complex operations. Here Altman’s warning against handing over control becomes an obligatory regulatory requirement; development teams, banks and government institutions in the region must build a technical architecture that keeps human oversight at the heart of the decision-making chain, protecting data and ensuring financial and legal audit, rather than blindly relying on outputs from models that consume billions of tokens.

Altman’s decisive stance in Abu Dhabi reveals that the new-generation battle is fought between two opposing forces: inference efficiency that outperforms the elite of researchers and requires unprecedented energy consumption, and strict engineering that refuses to surrender human leadership to the machine. Those who succeed in balancing the operation of agent swarms with preserving human intervention will be the ones who truly benefit from the models in sensitive work environments.