Tell us what you're trying to get done — we reply within one business day.
Not a dashboard — a plane every AI request physically passes through: policy checked, PII masked, output verified, action approved, cost metered, everything ledgered.
AI governance is the enforcement of policy, security, compliance, and accountability on AI systems. In SphereIQ it runs in the request path: every model call and agent action is policy-checked, PII-masked, verified for groundedness, approved where required, cost-metered, and written to an immutable audit ledger.
Declarative policies — data scopes, allowed models, allowed actions, approval thresholds — evaluated on every single call.
PII detection and masking, prompt-injection defense, output filtering, and groundedness scoring against Company Brain citations.
Answers checked against their cited sources before delivery. Low-confidence outputs get flagged, blocked, or routed to a human — per policy.
EU AI Act mapping, NIST AI RMF, ISO 42001 alignment, GxP validation documentation, and CSRD reporting — maintained as regulation moves.
Every request, decision, approval, and action — append-only, exportable, audit-grade. The answer to “show me exactly what the AI did.”
Cost metering per agent, team, and workflow, with ROI attribution in the Executive Console. The line item your CFO renews on.
Catches instructions planted in the input to override policy.
Catches attempts to talk the model out of its guardrails.
Catches attempts to make the model reveal its system prompt.
Catches attempts to move data where it should never go.
Catches attempts to reassign the model's role or authority.
Catches instructions hidden in retrieved content, not typed by a user.
Plus a secret-detector set covering API keys, tokens, PEM keys, JWTs and connection strings, redacted at the wire with traceable markers.
Deterministic: the same input always produces the same decision.
One audit ledger for chat, retrievals, policy hits, firewall events, governed actions, and agent publishes.
Verifiable against a published public key; e-Discovery export included.
Receipt shown as rendered by the platform · values illustrative
Tamper-evident, Ed25519-signed receipts (payload hash + signature + key fingerprint).
prohibited / high / limited / minimal / GPAI
A built-in classifier maps systems to risk tiers, with Annex III matching and per-article provider & deployer checklists, plus AI-literacy tracking.
Annex III matching · Per-article provider & deployer checklists · AI-literacy tracking
IQ/OQ/PQ
IQ/OQ/PQ validation protocols and a 21 CFR Part 11 control map, exportable as a reproducible validation package.
IQ/OQ/PQ validation protocols · 21 CFR Part 11 control map · Reproducible validation package
Of AI traffic passes the Governance plane — including third-party agents using SphereIQ tools via MCP
Deployed in validated life-sciences environments — the strictest compliance bar in the industry
Every AI dollar metered to an agent, a team, and an outcome via AI Economics Intelligence™
A fixed-scope engagement that inventories every AI tool, contract, and shadow deployment in your organization — then attributes cost, risk, and overlap, with a governed consolidation plan.
$8,500. Two weeks. Every AI dollar and risk, on one page.
Delivered by Sphere's architects. The output typically funds itself in the first consolidation it identifies — and it doubles as your EU AI Act inventory baseline.
Fixed scope
$8,500
2 weeks
Remote-first
Yes, progressively. Anything routed through SphereIQ’s model gateway or MCP hub is fully governed. The AI Spend Diagnostic inventories the rest, and the consolidation plan brings shadow tools into the governed path over time.
Policy evaluation adds milliseconds, not meetings. What it removes is the slow part: months of security review per AI project, because the controls are already in the platform and already evidenced in the ledger.
Policies define approval thresholds — by action type, dollar amount, data sensitivity, or confidence score. Approvals arrive where people work (Slack, Teams, email), and every decision is ledgered with the approver, timestamp, and context.
Read the chapters in any order. Deploy them in this one.