Ethotechnics Institute · Open commons
Make high-stakes AI easier to stop, explain, appeal, and repair
Australia's Robodebt scheme used software to raise about 470,000 debts the law did not allow. People appealed and won, one case at a time, and nothing changed for anyone else — so it ran for three years until a court stopped it. This site exists to make that failure harder to run.
Other frameworks ask whether you managed risk responsibly. This one asks whether the system can still be stopped while it is harming someone, and whether the people carrying that harm can make it stop. How the method works → Five public failures, scored → Why the laws hold →
The method, in six questions
Six questions to keep answering when software decides.
Software decides who gets a loan, a shift, a benefit, or a refund. A system becomes unsafe when these six drift apart — each card names the failure it catches.
- Capability What can the system actually do?
- Authority What is it allowed to do, for whom, and until when?
- Evidence What evidence justifies that permission?
- Dependency How hard is it to switch off or replace?
- Standing Who can challenge a decision, and with what force?
- Correction What can still be fixed, and how fast?
Each of these maps to one of the twelve laws. Read the method →
Incident triage
What is going wrong right now?
Choose the closest failure mode. Each path starts with what to do now, then names who should own follow-up and what evidence to collect.
- Decision appealed High · Owner in 24h
- Model made a harmful error Critical · Pause authority needed
- Appeals or reviews are stuck High · Escalate in 4h
- A user was harmed Critical · Human response now
- No owner is assigned High · Assign owner
- Decision cannot be clearly explained Medium · Review evidence
- System cannot be quickly stopped or rolled back Critical · Activate hard halt
Use these when a real person, queue, or decision needs attention—not just documentation.
The claim, in one figure
A delegation can double without anyone deciding that it should.
Both grants below grow by the same two percent a month. One is reviewed only when a single step is large enough to notice, so it is never reviewed. The other treats every widening as a fresh authorization, so it can be read, questioned, and reversed. That difference is what the standards on this site are for.
Demonstration
The same growth, decided and undecided
Two grants widen by 2% a month for 3 years. The left one is reviewed when a single step reaches 5%; the right one records every widening as an authorization.
The right-hand grant's state_history, first three entries
of 37
-
{ from: none, to: allowed, reason: issued }scope ×1.00 -
{ from: allowed, to: allowed, reason: expansion }+2%, scope ×1.02, capacity re-checked -
{ from: allowed, to: allowed, reason: expansion }+2%, scope ×1.04, capacity re-checked
A demonstration, not a measurement of any deployment. The steps and the
threshold are in src/utils/ratchet.ts; the transition reason
expansion and the states are the ones
authority-grant.schema.json allows, and a test holds them there.
MEC-19 expansion review reads state_history, and the left-hand
grant gives it nothing to read.
This is Law XI of twelve. Read the laws → See what STD-08 requires →
How it's different
Existing standards ask: 'Did you manage risk?' Ethotechnics asks: 'Can this system be stopped when it's harming someone?'
Not a replacement for NIST, OECD, or the EU AI Act. The enforcement layer they leave undertreated.
Most AI governance frameworks improve documentation, oversight, and accountability. Ethotechnics addresses what they leave undertreated: whether a running system can actually be halted, reversed, and repaired under stress — on a clock that matters to the person being harmed.
When a system simplifies its own operations, the work it leaves undone falls on someone — the nurse re-routing a scheduler's misfires, the claimant re-proving a denial. That labor is part of the system, so the person doing it holds standing to correct it (Law VII). The burden should run uphill.
| Existing standard says | Ethotechnics requires |
|---|---|
|
"Maintain human oversight" EU AI Act, Art. 14 |
Named human with stop authority, tested halt path, recovery clock |
|
"Manage risks across the AI lifecycle" NIST AI RMF |
Measurable time-in-harm bounds, exercised rollback and restoration paths |
|
"Conduct conformity assessment" ISO/IEC 42001 |
Evidence that the system can be stopped mid-incident, not just documented as compliant |
|
"Implement responsible AI principles" OECD AI Principles |
Binding escalation: owner + timer + action, or the system degrades |
Ethotechnics treats governability as an operational capability with measurable performance properties — not a compliance posture. See the full standards comparison for the detailed analysis.
Use and cite this work
Free to use with attribution
Ethotechnics Institute materials are published under CC BY 4.0 .
Individual entries and patterns include citation metadata so you can reference exactly what you used.
New and updated
Recent releases
-
Glossary v1.7.0
·
Added Non-Finality, Consequential Contradiction, and The Excluded Case, and generalized the falsified denominator into the falsified ledger — completing the five concerns and the fallibilist ground they stand on.
-
Diagnostics v1.3.0
·
Added the Corrective Debt Calculator, which prices the gap between action capacity and corrective capacity on two axes, with the absorption share and the workaround presumption as inputs.
-
Evals v1.5.0
·
Adds Corrective Learning (draft): whether the exceptions an institution handles change the machinery that produced them — exception absorption versus exception learning, the workaround presumption, and corrective debt. 15 eval suites, 148 test cases.
Studio
Need hands-on help?
Ethotechnics Studio provides commissioned support for clinical AI safety evaluation, diligence, governance design, and implementation work.