Option 3 · The Lines — what we defend, what we fear
Three thresholds, three different kinds of thing. Two of them are not dates at all — so this page refuses to draw them as dates.
AI under human control
The state this project defends: AI agents that are identity-bound (tied to a known, answerable person), envelope-bounded (limited in what they may do), and answerable (a human can be asked, and can answer, for what they did). It is not an event and has no date — it is a condition that either holds or fails, every day.
Status: holding today. Honest limit: a condition needs a meter or it stays aspirational — defining that meter is part of the student work.
Agentic collapse
The warned event: human control of the internet collapses under autonomous agent swarms — agents transacting with agents, at machine speed, beyond anyone's ability to supervise. Anthropic's Dario Amodei has warned an event of this kind could arrive within months. It is explicitly not AGI — naming it separately keeps the two from being confused.
Status: has not happened. It is an event, not a date — no year is claimed for it, and it will be detectable if it comes.
The WJC estimate, ~2032
The only true date of the three: the college's stake in the ground for general AI, set at the live consensus center (surveys + forecasting markets + capability curves — see the methodology). It is a stake, not data, and it moves when the evidence moves, at the quarterly re-cut, with reasons published.
Status: set ~2032; next scheduled re-cut January 2027.
Amodei warns about the red line. This project defines the yellow line. The students set the green.