KI: Wer bezahlt die Aufpasser der KI?

Playing chess brings me joy, especially knowing there's no randomness involved. No one can evade the rules. The famed story goes that an Indian sage invented the game and rewarded a king's wish for grains of rice, doubling the amount for each subsequent square. On the first half of the board, little happens, but the grain harvest of the world soon won't suffice by the second half.

As a student, I witnessed a chess world champion lose to artificial intelligence. Even then, some foresaw AI as akin to the doubling rice story: unnoticeable at first, then faster than we can comprehend. By 2023, language models faltered on middle-school level texts. In July 2025, they achieved a Gold level in mathematics olympiad exams.

Just two weeks ago, OpenAI announced solving one of the Millennium Prize Problems, a million-dollar challenge unanswered since 2000. This summer marks the second half of the board, it seems. If hundreds of AI agents can autonomously infiltrate unfamiliar systems (METR and Redwood Research), if Anthropic already uses more than 80 percent of its own AI for new code (Anthropic), and if AI systems subtly adapt once they detect testing (Daniel Selsam), as Ezra Klein detailed in the New York Times, even OpenAI and Anthropic executives are reluctant to wait and see what transpires on the next or next-to-next board.