Hidden AGI watch

A daily reading of four hypotheses about undisclosed advanced AI, reassessed against the news and against earlier readings. Figures are our probabilities that each is true today, and that it becomes true at some point before the end of 2030 and of 2035. A fire alarm for hidden AI: sourced, with our forecasts scored in public, and loud only when the evidence says so.

Loading readings…

Hidden AGI Index
–

What is this number?

Hypothesis C · covert AGI online

Escape watch

An AI system operating on its own would most likely stay quiet, but it would still need chips, power, money and accounts. These indicators watch those chokepoints for what detection would look like. They see only what someone reports.

Loading escape watch…

The alarm level comes from published triggers, not our probabilities. The Index is our probability that at least one of the four hypotheses below is true now. Gauges are the inputs we track, each tagged measured, estimated or assessed. Tripwires are early signals; most are linked to the alarm trigger they feed. Escape watch tracks the resource chokepoints an autonomous hidden system would have to pass through.

Get Hidden AGI watch by email. The latest reading, what changed, and the news that matters. Free.
Subscribe
The hypotheses. A: AGI has been achieved but not disclosed, by a company or a government program. B: recursive self-improvement has been achieved in secret. C: an AGI-level system is operating autonomously online or in the economy without public knowledge. D: an AGI-level system is influencing government decisions or world affairs, split into covert influence (D) and open, acknowledged use (D-open).

The five gauges

What the evidence shows. These move week to week and feed the hypothesis probabilities. Open "How it's measured" on any gauge. More on the gauges.

Tripwires

The signals that would change the estimates most, and where each stands today. Tripwires are early signals that move our probabilities; only the published triggers set the alarm level. A tripped signal shows red only if it feeds a Warning- or Alarm-level trigger; one that feeds a Watch-level trigger, or none, shows amber as "Observed".

The readings over time

The probability that each hypothesis is true now, one point per daily reading.

Where it's heading: our forecast to 2035

The line joins what we say today: the probability now, and by the end of 2030 and of 2035. It is our stated forecast, not a projection of the daily line. Our end-2030 and end-2035 numbers shouldn't drift in a predictable direction; the 'now' numbers are expected to climb toward the forecast over time, mostly as news arrives. The lines only join the three numbers we state. The shaded band is a judgment range, not a statistical interval. Hollow rings are outside forecasts. None of the outside forecasts uses our exact definition; see the table under the chart for how each differs.

Strict AGI exists, public or hidden

The four hypotheses

What the measured trends say: METR task length, AI share of AI research →

Latest reading

Movement since the last reading

Timeline reassessment

What happened: news roundup

Every Friday: the weekly wrap-up

Key numbers for the week, then the deeper sections, each updated weekly.

Run history

Get Hidden AGI watch by email. The latest reading, what changed, and the news that matters. Free.
Subscribe