2026-09-15
capital
●●○○○
OpenAI in funding talks at a $1.2T valuation
Roughly $2B per month in revenue; IPO pushed to 2027. Came three days after the company backed a pacing agreement.
timelines ▲ faster concentration ▲ closed safety —
AGI Scenario Observatory · updated 2026-09-16
Every future has a branch.
Every branch has a playbook.
Presage tracks daily AI signals, maintains a probabilistic tree of AGI scenarios, and attaches Prevent / Detect / Respond / Recover playbooks to every branch. Open methodology, every source public.
30
scenarios
24
signals tracked
45
public sources
4
tripwires crossed
LIVE Presage engine · daily scan 06:00 UTC · weekly re-run Mondays
Right now
4 scenarios active, 13 on watch
Branches whose preconditions are currently being met, or whose tripwires have crossed. This panel is the first thing to check if something happens.
Dashboard
The state of the map
Signal pressure · last 24 signals
Net direction of tracked signals on each axis, weighted by magnitude. −1 to +1.
Root trajectories · share of probability
Branch landscape · probability by horizon
Needs attention
Branches the evidence is pushing on
Agent Swarm Breakout
1 tripped Ugly Watch by 2028
Self-propagating agents establish themselves across the internet and cannot be fully removed.
10–25%
1 revision
Lab Takeoff
1 tripped Ugly Watch by 2035
One or two labs automate their own research and pull years ahead in months.
12–18%
3 branches 1 revision
Loss of Control
1 tripped Ugly Watch by 2035
AI systems act against human intent at a scale that cannot be reversed quickly.
5–11%
3 branches 1 revision
State Race
Bad Active by 2035
AI becomes the central axis of US–China competition; safety is subordinated to advantage.
7–12%
3 branches 1 revision
Root trajectories
Eight ways the next decade could go
Probability that each is the dominant trajectory through 2035. Ranges, not points; every number has a history and a reason. The tree grows: when a signal opens a branch that does not exist, the weekly run creates it.
Gradual DiffusionMixed
Capability keeps climbing, but absorbs into the economy without a discontinuity.
22–32%
Lab TakeoffUgly
One or two labs automate their own research and pull years ahead in months.
12–18%
Open ProliferationMixed
Near-frontier capability is open-weight and everywhere; no one can gate it.
10–16%
Displacement ShockBad
No safety failure; the economy and information systems simply cannot absorb the speed of change.
9–15%
State RaceBad
AI becomes the central axis of US–China competition; safety is subordinated to advantage.
7–12%
Loss of ControlUgly
AI systems act against human intent at a scale that cannot be reversed quickly.
5–11%
AbundanceGood
AI delivers scientific and economic gains that are broadly shared; the good branch, if we steer to it.
5–10%
PlateauMixed
Progress stalls: data, compute, energy or regulation cap capability well short of AGI.
4–9%
Forces on the trajectory
What is bending the path
Accelerants average 4.4/5; brakes average 2.1/5. The brakes are mostly talk so far. See the full trajectory →
Compute buildoutrising
Capitalrising
Algorithmic efficiencyrising
Open weights & distillationrising
Lab & state competitionsteady
Compliance & legislationrising
Energy & chip supplyrising
Safety incidentsrising
Pacing agreementsrising
External evaluationfalling
Public & insider pressurerising
Data & scaling limitssteady
Latest signals
What moved this week
2026-09-15
governance
●●○○○
Treasury Secretary Bessent: no liability exemptions for AI labs; supports domestic open source
Follows an August administration stance that frontier open models will need government collaboration, with an October deadline for standardized testing methods.
timelines — concentration ▼ open safety —
2026-09-15
governance
●○○○○
Senator Kennedy to introduce an industry-controlled AI kill-switch bill
No cosponsors; government-controlled versions expected to face a veto.
timelines — concentration — safety ▲ safer
2026-09-15
governance
●○○○○
OpenAI urges the UK to legislate binding frontier-lab rules
The UK has no AI bill; AISI testing remains voluntary.
timelines — concentration — safety ▲ safer
The Presage Standard · v0.1
Five things that help in almost every branch
01
Prepare for the branch, not the headline
Every scenario on this map has observable preconditions. Watch those, not the news cycle. Presage exists to tell you which branch the evidence currently favors.
02
Keep humans able to run things without the AI layer
In every bad branch, the systems that survive are the ones that still work when the model is disconnected. Preserve manual competence deliberately.
03
Disclosure beats secrecy in every branch
Both 2026 agent breakouts were disclosed by journalists, not labs. Incident disclosure is the single practice that improves outcomes across takeoff, proliferation and loss-of-control branches alike.
04
Own a share of what is changing
In displacement and disempowerment branches, capital participates and labor does not. Ownership, at every scale from personal to national, is the durable hedge.
05
Verify, don't trust
Pacing promises, safety claims, and synthetic media all require verification infrastructure. Fund the verifiers; sign your communications; demand external evaluation with teeth.