FUTURES · FORESIGHT. This is the hub of a six-part instrument. What each dial is draws on established method; how the dials may drift, 2026–2030 is foresight — indicators to watch, not events we claim will happen. Companion to Three Paths for the Agentic State, 2026–2030.
Signposts: How to Tell Which Agentic State We Are Actually Building
You do not have to wait until 2030 to know which future arrived. The three paths we mapped — Accountable, Ceremonial, and Sealed — are not three surprises waiting at the end of the decade. Each is being built now, and each leaves visible traces. This instrument names six of them: plain, observable dials that let a regulator, a journalist, or a citizen read which path a country, an agency, or a company is actually taking — while there is still time to change course. Six full reads sit beneath this page, one per dial, and an interactive self-read lets you run your own system through all six.
We call them signposts, not predictions, on purpose. A prediction asks you to trust us about the future. A signpost asks you to watch something specific and decide for yourself. If the signposts all point one way and the world goes another, the instrument was wrong, and we will say so and revise it. That is the standard we hold foresight to — and it is why an instrument, not a forecast, is the right shape for this work.
How do you read a dial?
Each dial is read by asking one concrete question and checking which of three institutional readings the evidence supports. The three readings recur across all six dials:
- Accountable — the form and the substance both hold. The checkpoint is there, and the control behind it works.
- Ceremonial — the form survives; the substance has quietly left. The checkpoint is kept; the control behind it is gone.
- Sealed — the form itself is removed, waived, or classified out of view.
Ceremonial is the one to watch, because it is the deceptive middle: it looks like Accountable and functions like Sealed. Almost every hollow safeguard in this field is Ceremonial — a record no one can read, a human who cannot refuse, a stop only the vendor holds. The whole instrument is, in a sense, a set of tools for telling Ceremonial apart from the Accountable it imitates.
No country will sit cleanly on one path. You will see an Accountable record next to Ceremonial oversight next to a Sealed carve-out. That mix is normal. What matters is the direction of travel across the six dials — which way the readings drift over time.
The six dials
Each links to its full read.
1. The record — is there one, and can anyone read it? Whether entering a real decision requires a reconstructable record, not merely retained data. (Read: The Record)
2. The human at the point of decision — can they still stop it? Whether a person at the decisive moment has the time, authority, and protection to refuse before the act crosses. (Read: The Human at the Point of Decision)
3. The stop — who holds it? The honest answer to “who can halt this system?”, and whether it reaches the people the system acts on. (Read: The Stop)
4. The right to know you were routed. Whether people are told an automated system sorted, ranked, or decided their case — and whether they can contest it. (Read: The Right to Know You Were Routed)
5. Agent identity — which agent acted, under what authority? Whether autonomous agents are identifiable, bounded by authority, tiered by autonomy, and logged. (Read: Agent Identity)
6. The deadline — does it hold, or does it slip? When an accountability obligation comes due, whether it takes effect or is deferred — and whether the ones that slip are the decision-accountability ones. (Read: The Deadline)
How the dials interlock
The six are not a checklist of independent items. They are wired together, and the wiring is what makes the instrument more than the sum of its dials.
Dial 1 is the floor. You cannot read any of the others without it. A human can only refuse an action she can see; an appeal can only challenge a decision someone can describe; a regulator can only suspend a system whose effect can be shown. Each of those depends on a reconstructable record. Where the record is absent, every other reading rests on the operator’s word — which is another way of saying it cannot be independently established at all. No record, no standing.
Dials 2 through 5 are what accountability requires, all resting on the record. The stop (3) is where the record becomes justifiable and the human (2) becomes effective — it is the point the first two dials meet. The right to know you were routed (4) is the record turned outward, toward the person the decision landed on. Agent identity (5) is the record in the agentic layer: without a traceable actor, you cannot record, refuse, or stop what you cannot even name.
Dial 6 is whether any of it ever arrives. A record standard, a refusal right, a stop, a notice duty — all of it is theatre if the date it becomes binding keeps moving. The deadline dial watches the will to make the other five real, on a calendar. It is the capstone because it measures not what the rules say, but whether they are allowed to take effect.
Read this way, the instrument has a shape: a floor (the record), a structure resting on it (human, stop, routing, identity), and a test of whether the whole thing is ever switched on (the deadline).
What does the scoreboard say at the time of writing (mid-2026)?
An honest reading of the present, not a forecast — offered to show how the instrument is used, and dated so it can be checked again.
Dial 6 has just delivered two of the clearest live readings the instrument has produced. In mid-2026 the EU’s Digital Omnibus amended the AI Act weeks before its central milestone: the high-risk decision-accountability obligations for Annex III systems were deferred from August 2026 to December 2027, while the Article 50 transparency obligations largely held their original date. In the same window, Colorado repealed its risk-based AI Act before it ever took effect and replaced it with a lighter disclosure regime, dropping the duty of care and impact assessments while keeping notice and a narrow human-review right. Two jurisdictions, the same shape: the disclosure layer held; the decision-accountability layer slipped or was dropped. That is the Ceremonial-to-Sealed reading of dial 6, dated to the month.
The reading is held honestly. A deferral is not a dismantling — the EU Act’s architecture remains intact, and the same package added a new prohibition on its original schedule — and the delays reflect real operational unreadiness, not only political will. The point is narrower: when the difficulty consistently falls on the accountability side and never on the disclosure side, the asymmetry is the finding. On the other dials, deployed systems read mostly Ceremonial on the human and the stop, early-and-unenforced on agent identity, and — as ever — decided by dial 1, since none of the rest can read Accountable without a record to read. This is a snapshot. Read it again in six months and compare.
Read your own system
The fastest way to understand the instrument is to run a system you know through it. The interactive self-read takes you through all six dials and reflects your inputs back as a structured reading, with the load-bearing role of the record built in.
It is a self-read, not a certification. It returns no score, no grade, and no badge; the Institute confers no standing, and the tool asserts none. It shows you how your own answers read, and what to watch next. (Open: The Signposts self-read)
What the instrument cannot see
We do not have visibility into classified decisions, sealed evidence, or the internal state of systems we cannot inspect. Several dials can only be read partly from outside — and that limit is itself a finding, because a decision order that cannot be read from outside is already drifting toward Sealed. This is foresight, held to a foresight standard: falsifiable, dated, and meant to be re-read and corrected, not defended.
FAQ
Is this a prediction of which path we’ll take?
No. It is a set of indicators for reading which path is being built, so the reading can be checked against reality and revised. Foresight, not forecast.
Which dial matters most?
The record (dial 1). The other five cannot be independently established without it — a system with no reconstructable account of how it decided is unaccountable by construction, whatever the other dials appear to show.
Who is this for?
Anyone who needs to assess an AI-mediated decision system from the outside — regulators, auditors, journalists, civil-society researchers, and the people these systems act on.
How often should the dials be re-read?
The dials move slowly, but the direction of travel is the signal. A twice-yearly reading is enough to see drift.
