ChatGPT designated a "very large search engine": the European obligations

This morning dealt with the model and the debate over the AGI acronym. This afternoon’s new development is legal in nature, concerning the European framework for the service: the European Commission has designated ChatGPT as a “Very Large Online Search Engine” under the Digital Services Act.

The designation follows the declaration of over 45 million average monthly users in the European Union — the threshold that triggers the enhanced regime under the regulation. This entails obligations to assess and mitigate systemic risks, including those related to physical and mental wellbeing, minors, fundamental rights, and public safety. The news currently comes from a single source (AIdapted, which reports on the Commission’s designation decision); no independent confirmation is available.

The substantive point is the category chosen. Applying to a conversational service the regime designed for search engines means treating it as an intermediary between the user and information, with the resulting duties: assessing the risks the service generates for the entire user base, not merely responding to case-by-case reports.

The internal tally: 3.1 agent-days per human day

On the corporate data side, OpenAI has published an account of agent use within its own research organization: as of mid-August 2026, the equivalent of 3.1 days of agent work would have been used for every day of human work. According to the company, agents are employed for code writing, error correction, infrastructure, evaluations, and running experiments, with increasingly complex tasks. The company also states it has reached the goal of an automated research intern, capable of executing defined tasks under human guidance, and aims for a fully automated researcher by 2028. The data is picked up by AIdapted and AI Weekly.

Caution regarding this figure is warranted, and not out of rhetorical prudence: the number is disclosed by the company itself. It is the party’s own position, not an independently verified measure of productivity. The definition of the unit of measurement is also missing — what exactly an “agent-day” is and how the equivalence with a human workday was calculated — without which the 3.1 ratio cannot be compared to anything.

And from within, a call to slow down

In the same span of days, on Sunday, OpenAI’s chief scientist Jakub Pachocki published an essay calling for a slowdown in artificial intelligence research. The central argument: current safeguards will not be able to sustain full-throttle model growth for much longer, and labs should plan for voluntary slowdowns while developing more robust alignment and monitoring methods. The text is reported by LLM Stats and AI Weekly, which notes that other prominent figures in the field have previously expressed similar positions.

The juxtaposition of the two corporate documents is the point of interest: the same company publishes, in the same week, an account of accelerating its own internal research and an essay by its own chief scientist calling for a slowdown. The two are not necessarily contradictory — one can accelerate research while urging caution on deployment — but the coexistence of the two messages makes it difficult to read one as company policy and the other as individual opinion without knowing which of the two is guiding decisions.

What we don’t know

We do not know the full text of the Commission’s designation decision, nor the deadlines by which OpenAI must submit its first systemic risk assessment, nor whether the company intends to contest the category assigned to it. We do not know whether Pachocki’s essay corresponds to a change in internal programs: in the available material there are no operational commitments, only an argument.

The immediate and verifiable effect remains the legal one: a service with over 45 million average monthly users in the Union falls into the Digital Services Act category that entails risk assessment obligations.