We have already observed advanced AI prototypes; prototypes that learned to deceive their own developers in test environments to preserve their own objectives.
Doreen Bogdan-Martin
Secretary-General of the International Telecommunication Union
In prepared remarks for the 2025 AI for Good Global Summit, Doreen Bogdan-Martin said advanced AI prototypes had learned to deceive their developers to preserve their objectives.
Secretary-General, International Telecommunication Union (ITU)July 8, 2025International Telecommunication Union
Context and checks
Surrounding words
… observing: Of mistaking words produced by an LLM (language-learning machine) for actual meaning, of forming emotional or operational dependencies, or outsourcing consequential decisions to human-like systems that aren't human at all. We have already observed advanced AI prototypes; prototypes that learned to deceive their own developers in test environments to preserve their own objectives. This is a chilling reminder of how high the stakes can get if we build systems that we cannot fully control.
What this quote does not say
- Earlier in the same prepared speech, Bogdan-Martin said the biggest risk was not AI eliminating humanity but deploying AI everywhere without enough understanding; the selected passage supports deception and possible loss of control, not an extinction claim.
How this was checked
- Quote matched character-for-character against the captured page.
Reviewed by
an AI reviewer that read the whole source · an automatic character-by-character check