A diverse group of young colleagues smiling together around laptops, including a wheelchair user, a person with blue hair, and a person wearing a hijab, beneath posters reading all voices shape the future and diversity equity inclusion belonging.

GNOSIS Ethical Intelligence · signed · sealed · provable

Every AI answer
needs a witness✱

Ask an AI to prove what it just said, and it hands you a link. A link is not proof. It came from the same model that wrote the answer.

A witnessed answer is different: the source, dated. The human, named. The record, locked. And when an answer cannot be proven, a person steps in.

Soon, every serious AI will be required to prove what it says. This is where answers come to get their seal.

See it fire Read the thesis

The proof layer any AI can plug into Built in Newfoundland and Labrador, for the edges

01 · Inclusion is the architecture, not the afterthought

Design from the edge, and the middle comes free.

The curb cut was built for the wheelchair. It ended up serving the parent with a stroller, the traveller with a suitcase, the worker with a cart. Accessibility, done at the source, is a gift to everyone.

GNOSIS is the same idea moved into the accountability layer. It does not matter whether the person on the other end is gifted, average, neurodivergent, Deaf, or living with a disability. Everyone gets the same verifiable record. But the system is tuned, on purpose, to the people the average was never built to hold, because if it holds for them, it holds for all of us.

The most honest thing a system can say to an outlier is "I cannot verify this." Systems that pretend to certainty force the people at the edges into false categories. GNOSIS refuses to guess. That refusal is inclusion, written into the machine.

02 · Mechanism

Detect. Decide. Document.

One loop runs on every AI-assisted output and resolves to one of three verdicts: VERIFIED, REFUTED, or UNVERIFIABLE. The record it leaves is signed and append-only.

Detect

Intercept the output

Every AI-assisted output is captured before it is used, with the model, prompt context, and claim attached. Nothing reaches a record unmarked.

Decide

Check, then resolve

The output is checked against a dated, verifiable source and a named responsible human, then resolves to a verdict. Low-confidence cases are never forced into a false positive.

Document

Sign and append

The result is written to an HMAC-signed, append-only ledger. Human-in-the-loop is preserved as a record, not a claim. There is no delete key.

UNVERIFIABLE · the differentiator

When a claim cannot be traced to a source, GNOSIS does not guess. It returns UNVERIFIABLE and routes the case to a human, instead of emitting a confident answer that is wrong. This is the sharpest guarantee in the system: the layer would rather say it cannot verify than certify something it cannot stand behind.

03 · When it fires

This is what it means
when it fires.

Four moments from the software we run today. Four people who will never read a line of code. The same engine sits underneath every one of them, deciding what can be proven, and refusing to guess.

CIVICHuman, now

A wildfire is moving toward town. A Deaf resident calls for help, and an AI is triaging the call as the network starts to fail.

interpreter queued · works offline · delivered on reconnect · gn_88409 · signed

GNOSIS Ethical HealthRefuted

A doctor's AI scribe writes 5 mg into a medically fragile child's chart. The formulary says that dose could hurt her.

source: Formulary 2024, p.212 · blocked before the chart · gn_88411 · signed

InclusionWorksUnverifiable

An AI report is one sentence away from labelling a grade 4 student below reading level for the rest of the year.

no source fits this learner · routed to a human · gn_88412 · signed

KARATUnverifiable

A shipment clears customs with clean paperwork. The paperwork cannot say whose small hands packed the boxes.

provenance gap at tier 3 · shipment flagged · gn_88408 · signed

Every one of those catches came from the same engine. Under the hood, it looks like this.

gnosis · POST /v1/notarize → record

Verdict

how a system calls the layer 
# any model, any framework
POST https://api.gnosis.eco/v1/notarize
Authorization: Bearer sk_live_…
{
  "model": "your-model/v3",
  "output": "…generated text…",
  "context": { "domain": "clinical" },
  "on_unverifiable": "route_to_human"
}

→ 200 { "verdict": "UNVERIFIABLE", "routed_to": "human", "record": "gn_c7e19a3f" }

If your AI touches a person, a chart, a classroom, an emergency call, or a supply chain, it needs this engine under its hood. Anyone using AI needs the notary.

04 · The thesis

Accountability is becoming a requirement, not a feature.

AI can cite. It cannot prove. Those are different businesses.

Everything that matters has a notary. Land, wills, marriages, companies: none of them run on trust alone, they run on a neutral witness who stamps the record. AI has never had one. GNOSIS is that missing notary. It binds each AI output to a dated source anyone can check, a named human who stood behind it, and a record locked against quiet edits. It works with any model. Any system can call it.

"But AI links its sources now." It does. Here is what the science says about those links.

51.5%

Barely half of the sentences produced by citation-backed AI search engines were fully supported by the sources they cited.

Liu, Zhang & Liang · Stanford · 2023
58-82%

Hallucination rates when general chatbots answered legal questions, with fabricated or misgrounded authorities.

Stanford RegLab & HAI · 2024
38%

Share of webpages that existed in 2013 and were simply gone a decade later. A link is a live address, not a frozen fact.

Pew Research Center · 2024
0

Audit systems, in any regulated industry, that accept self-attestation. The party that produces a claim never gets to be the party that certifies it.

Standard audit independence principle

The pattern under all four numbers is the same: a link is the model grading its own homework. The citation is generated by the same process that generated the answer, so it inherits the same failure modes, and it points at a live page that can change or vanish after the fact.

GNOSIS is built on the opposite principle, the one audit science settled long ago: verification must be independent of generation. A separate layer checks the claim against a dated snapshot of the source, a named human signs the outcome, the record is locked append-only, and when the check cannot be completed the system returns UNVERIFIABLE instead of a confident guess. None of that is an opinion about how good the model is. It is a record of what was checked, by whom, against what, and when.

05 · One engine, two streams

The software, and the layer underneath.

Stream 01 · Our software

Built on the engine, for the edges

Four products run on the layer today, in the hardest real conditions we could find. Every screen they show is backed by a signed record.

  • CIVIC · emergency alerts that stay accountable and keep working offline.
  • GNOSIS Ethical Health · catches AI documentation and supply-chain errors before they reach a medically fragile patient.
  • InclusionWorks · accountable AI for learners the average system leaves out.
  • KARAT · provenance for supply chains, from source to shelf.
Stream 02 · The layer, for sale

The same engine, as infrastructure

Model-agnostic and API-first. Any AI company can put the engine under its own hood: one call per output, a verdict back, a signed record on the ledger. Proven at the edges once. Sold as proof everywhere.

See the engine

06 · The workforce

The most documented generation in history is clocking in.

Gen Z and Gen Alpha arrive with documented needs, legal literacy, and expectations. They disclose, they ask, they report, and they expect to be taken seriously. Most managers were never equipped for the accommodation requests, mental health disclosures, and documentation these moments require. That is not a training gap. It is an infrastructure gap, and it is the one GNOSIS was built to close.

Baseline

Accommodation is the new baseline

This generation arrives with documented needs. The organizations that build real accommodation infrastructure will not just avoid harm. They will keep their best people.

Voice

Every report deserves a witness

When someone reports bullying, discrimination, or harassment, the response should be structured and held in a record, not left to memory.

Retention

Respect is a retention strategy

The workplaces that invest in structured support today will keep the workforce everyone else is struggling to hold on to.

2040

Canada has set the date: barrier-free by 2040 under the Accessible Canada Act. Between now and then, "we are inclusive" stops being a slogan and becomes a claim your business has to prove. GNOSIS keeps the notarized record: every accommodation request, every disclosure, every report, every response. Dated, named, and locked. Inclusion you can show, not just say.

07 · Standards

Optional is becoming required.

Standards bodies build the measuring stick. GNOSIS is the record that gets measured.

Frameworks like CAN-ASC-6.2 and the trusted-AI certification now taking shape are turning accountability from a feature into a requirement. Those criteria are being written right now, in Ottawa. Standards define the measuring stick; GNOSIS produces the record it measures: the dated sourcing, named responsibility, and tamper-evident ledger the certification will ask every AI to show. Two halves of one certification story, and the reason this layer belongs in the rooms where the criteria are written.

08 · The horizon

Every AI answer, notarized.

Not one accountable app. One notary, called by any system, under every AI output, stamping what can be proven and saying so when it cannot, built so the hardest case is the one it holds first.

09 · Integrate

Build on the accountability layer.

  • One API call, any model. Every output comes back notarized: a verdict, and a signed, dated record on the ledger.
  • Human review as evidence. On UNVERIFIABLE the case routes to a person, and the handoff is recorded, not asserted.
  • Model-agnostic, standards-aligned. The reference implementation the accountability frameworks imply.
Contact us See the proof

Write to us at info@gnosisethical.com. A person answers.