Human-Oversight Infrastructure · Patent Pending

Human oversight infrastructure for high-risk AI decisions.

COHESION saves humanity by keeping human judgment alive in the age of AI.

Evaluate AI risk (DRS), measure human judgment quality (JIS), and create replayable decision oversight across hiring, lending, healthcare, insurance, fraud, and eligibility workflows.

curl this curl https://api.cohesionauth.com/v1

Mapped to the regulators auditing your AI

  • Colorado SB 26-189 Effective 2027-01-01. AI consumer-protection mandate.
  • NIST AI RMF 1.0 Govern. Map. Measure. Manage. JIS feeds Measure.
  • SEC AI Disclosure Material AI-system risk to the registrant. DRS routing evidence.
  • White House OMB M-25-21 Rights-impacting + safety-impacting AI use cases.
  • EU AI Act Article 14 Human oversight, effective 2027-08-02 for high-risk systems.

A scoring API, not a slide deck.

50+ named endpoints. 80 ms timing floor. Tamper-evident HMAC-chained audit log. Methodology Annex PDF per organization, regenerated when the spec moves. The self-describe endpoint is a 2 KB JSON catalog of every public route.

  • Real-time JIS scoring POST `/v1/score` returns a 7-dimension Judgment Independence Score, a decay projection, a maintenance recommendation, and a HMAC chain entry. 80 ms timing floor enforced in the scoring engine.
  • DRS routing for AI decisions POST `/v1/decision/score` returns a Decision Risk Score, a routing recommendation (auto / must_review / policy_blocked), and a fail-closed contract on every non-2xx. The reviewer queue surfaces high-risk decisions.
  • Signed Methodology Annex per customer The PDF that lets your auditor say yes. Mapped to nine US frameworks plus four international. Regenerated quarterly on the Audited tier.
  • Drop-in SDKs @cohesionauth/sdk (npm) + cohesion-sdk (PyPI). v1.3.0 ships the cognition-probe surface, dual-phase telemetry hook, and full Coverage Lane support.
# Score a human-AI decision in 80 ms
curl https://api.cohesionauth.com/v1/score \
  -H "Authorization: Bearer $COHESION_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"operator_id":"op_alice","session_id":"sess_42","domain":"financial","interactions":[...]}'

# Response
{
  "jis": 82.4,
  "band": "proficient",
  "dimensions": { ...7 dimensions... },
  "decay_projection": { "days_until_review": 23 },
  "compliance": { "frameworks_mapped": ["NIST_AI_RMF", ...] }
}

Three tiers. One assurance ladder.

The score is identical across every tier. What differs is verification rigor, decision volume, and procurement-readiness. Full pricing details →

Starter (Self-Reported)
$25,000
/ year · 100K decisions / mo · 1 domain

One AI decision domain in production, defensible artifact for the next audit cycle.

Starter details
Audited (Pro)
$75,000+
/ year · 500K decisions / mo · 3 domains · quarterly refresh

Third-party-verifiable evidence. Big-4 compatible artifact bundle. Quarterly methodology refresh.

Audited details
Enterprise
$150,000+
/ year · custom volume · all 6 domains · custom MSA

Unified oversight posture across business units. Dedicated methodology liaison.

Enterprise details

Founding Design Partner Cohort

10 seats. Closes 2026-06-05. Co-design the certification your industry will be measured against. Apply »

Industries deploying COHESION today.

Six domains live in the API enum. Each has named regulatory pincer requirements that JIS + DRS map to directly.

Financial services

Underwriting, loan denial, fraud-detection AI. Mapped to SEC AI guidance + NYC LL 144 (when extended to credit). Forced-escalation rules enforce policy on every loan-denial decision.

Healthcare

Clinical-note AI, prior-auth AI, ED triage AI. FDA AI/ML SaMD framework. Forced escalation on any healthcare protected-class outcome.

Aviation

Maintenance-decision AI. FAA AI roadmap. Domain-specific decay thresholds for high-stakes pilot + maintenance roles.

Legal

Contract review, e-discovery, compliance triage AI. Audit-log export for matter retention. Whitelisted decision serialization for matter privilege.

Pharmaceutical

Suspicious activity reports, anti-counterfeit AI. FDA + SEC dual jurisdiction. Methodology Annex Big-4 co-sign at the Audited tier.

General

SOC-analyst triage, resume-screening AI (NYC LL 144), customer-service deflection AI. Default catch-all domain in the live API enum.

Score one decision in the next 90 minutes.

That is the typical first-pilot integration time. Drop in the SDK, point it at a webhook, see your first JIS for a real operator before the end of the call.