Assessment · Governance · Test and control

12Fairness, explainability and human oversight

Do people review what matters, and can we explain each decision?

Reviews that catch real errors, bias tests where decisions affect people, and decisions you can explain.

AutomatedStakeholder interviewsUses your existing assessments 2 controls 2 engines

How we collect

  • AutomatedReview rates and outcomes from workflow logsNeeds your export or setup
  • Stakeholder interviewsOwners: decisions that affect people, required sign-offsRuns today
  • Uses your existing assessmentsBias audits and model validation reportsNeeds your export or setup

Status labels show what runs today, what needs an export or setup from you, and what is planned.

What you get

  • Review coverage and catch rate
  • Decisions that need a person, by risk tier
  • Bias testing where the law requires it (planned engine)

Questions we ask

  • Which decisions must a person sign?

We can start from

Bias audits (NYC LL 144), model validation files

Data sources

SourceWhat we readStatus
QuestionnaireThe short interview for what data cannot show: purpose, owners, decisions AI informs, regulations in scope.Runs today
AI gateway telemetryLiteLLM, Cloudflare AI Gateway or AICost CostWall logs: cost per team, workload, agent and outcome.Partly automated

Built on governance engine modules

P12 PlannedQuality, Hallucination & Evaluation

How wrong is acceptable, per consequence class, and how do you know?

Security & Assurance
P19 PlannedVertical Governance

What does your industry add on top, HIPAA, legal privilege, finance?

Assess & Govern

AICost governance controls in this module

AICost-GovControl603Wave 2
Bias testing and explainability
What it is

Tests for unfair outcomes in AI used for hiring, lending, pricing or access to services, and the level of explanation each decision needs.

Why it is needed

Biased decisions create legal claims and damage trust. Some laws require explanations people can understand.

Why enterprises need it

HR, lending and customer teams need test results and explanations they can show a regulator or court.

What you get:Discrimination risk measuredExplanations matched to riskAudit trail for claims
Run by
AI Quality and Acceptance Gate engineRegulation, Audit and Compliance engine
Data
Questionnaire
Evidence produced
Bias test record: system, groups tested, metric, result, threshold, action, explanation method.
Maps to
EU AI Act: high-risk duties; NYC Local Law 144; Colorado AI Act
AICost-GovControl604Partly automated
Human oversight that works
What it is

Defined points where a person reviews or approves AI, with the authority and time to overrule it, and records of what they did.

Why it is needed

Reviewers who approve everything add cost without control. Oversight needs design, not just a step.

Why enterprises need it

Regulators ask for proof that oversight changes outcomes. Finance asks what review costs.

What you get:Review where it matters, not everywhereOverride rates trackedReview cost known
Run by
Policy Drafting engineHuman Review engine
Data
AI gateway telemetry, Questionnaire
Evidence produced
Oversight table: system, decision, reviewer role, approval rule, override rate, review cost per month.
Maps to
EU AI Act: human oversight; NIST AI RMF: Govern

Frameworks

Sector rulesHealthcare, financial model risk, employment AI laws

Stakeholders

HRComplianceModel risk
← 11. Security and red teaming All modules 13. Agent safety and runaway control →
📖 Data sources & methodology 172 text models · 9 embeddings · 44 vision · 62 audio · 8 vector DBs across 10 vendor pages · last verified 2026-10-18

Methodology

  • All prices are USD per 1 million tokens, current as of 2026-10-18.
  • Vendor-published values have no mark. Inferred/extrapolated values are marked with * and listed below.
  • Batch API discounts are 50% off standard rates across providers that offer Batch mode.
  • Prompt caching discounts vary by provider (typically 80-90% off cached input tokens).
  • Regional data-residency surcharges (Anthropic 1.1x, OpenAI 1.1x, Google regional tiers) are NOT included in base rates.
  • Long-context pricing tiers apply when input exceeds model threshold.
  • Embedding prices are input-only (no output tokens generated).

Primary sources

Last-verified date is the most recent successful daily snapshot (aicost_pricing_snapshots) or, when no snapshot exists yet, the latest successful crawler run (aicost_crawler_runs). 10 of 10 vendors are currently verified. Aggregator services (TokenCost, AI Pricing Guru, etc.) are not listed.

Anthropic
2026-10-18
https://www.anthropic.com/pricing
Daily snapshot since Sep 2023 · 704 days captured
Anthropic Docs
2026-10-18
https://platform.claude.com/docs/en/about-claude/pricing
Daily snapshot since Sep 2023 · 704 days captured
OpenAI
2026-10-18
https://openai.com/api/pricing/
Daily snapshot since Sep 2023 · 705 days captured
Google AI
2026-10-18
https://ai.google.dev/gemini-api/docs/pricing
Daily snapshot since Dec 2023 · 680 days captured
Google Vertex
2026-10-18
https://cloud.google.com/vertex-ai/generative-ai/pricing
Daily snapshot since Dec 2023 · 680 days captured
DeepSeek
2026-10-18
https://api-docs.deepseek.com/quick_start/pricing
Daily snapshot since May 2024 · 619 days captured
xAI
2026-10-18
https://x.ai/api
Daily snapshot since Nov 2024 · 537 days captured
Mistral
2026-10-18
https://mistral.ai/pricing
Daily snapshot since Dec 2023 · 678 days captured
Cohere
2026-10-18
https://cohere.com/pricing
Daily snapshot since Sep 2023 · 704 days captured
Voyage AI
2026-10-18
https://docs.voyageai.com/docs/pricing

Inferred values (marked with * in calculator tables)

Derived from industry conventions, not directly published by the vendor. Typical conventions: cached input = 10% of base (90% off), Batch API = 50% of base (50% off).

Vendor / Model Field Why it’s inferred
Anthropic — Claude Sonnet 4.6 cachedInput Derived at 10% of input rate — Anthropic publishes 90% cache-hit discount on this tier.
Anthropic — Claude Sonnet 4.5 cachedInput Derived at 10% of input rate; same 90% cache-hit convention as Sonnet 4.6.
Anthropic — Claude Sonnet 4.5 batchInput Derived at 50% of standard input — Anthropic documents uniform 50% Batch discount.
Anthropic — Claude Sonnet 4.5 batchOutput Derived at 50% of standard output — Anthropic documents uniform 50% Batch discount.
Anthropic — Claude Haiku 4.5 cachedInput Derived at 10% of input rate — Anthropic 90% cache-hit discount convention.
OpenAI — GPT-5.4 Mini cachedInput Derived at 10% of input — OpenAI documents automatic 90% discount on cache hits across GPT-5.x tier.
OpenAI — GPT-5.4 Nano cachedInput Derived at 10% of input — OpenAI 90% cache-hit convention.
OpenAI — GPT-5.4 Nano batchInput Derived at 50% of input — OpenAI Batch API uniform 50% discount.
OpenAI — GPT-5.4 Nano batchOutput Derived at 50% of output — OpenAI Batch API uniform 50% discount.
OpenAI — GPT-5.4 Pro cachedInput Derived at 10% of input — OpenAI 90% cache-hit convention.
OpenAI — GPT-5.4 Pro batchInput Derived at 50% of input — OpenAI Batch API uniform 50% discount.
OpenAI — GPT-5.4 Pro batchOutput Derived at 50% of output — OpenAI Batch API uniform 50% discount.
OpenAI — GPT-5.2 cachedInput Derived at 10% of input; no residency uplift.
OpenAI — GPT-5.2 batchInput Derived at 50% of input.
OpenAI — GPT-5.2 batchOutput Derived at 50% of output.
OpenAI — GPT-5 cachedInput Derived at 10% of input.
OpenAI — GPT-5 batchInput Derived at 50% of input.
OpenAI — GPT-5 batchOutput Derived at 50% of output.
OpenAI — GPT-5.5 Pro cachedInput Derived at 10% of input — OpenAI does not publish a cached rate for *-pro models; using the family convention.
OpenAI — GPT-5.5 Pro batchInput Derived at 50% of input.
OpenAI — GPT-5.5 Pro batchOutput Derived at 50% of output.
OpenAI — GPT-5.2 Pro cachedInput Derived at 10% of input — pro-tier convention.
OpenAI — GPT-5.2 Pro batchInput Derived at 50% of input.
OpenAI — GPT-5.2 Pro batchOutput Derived at 50% of output.
OpenAI — GPT-5.1 batchInput Derived at 50% of input.
OpenAI — GPT-5.1 batchOutput Derived at 50% of output.
OpenAI — GPT-5 Pro batchInput Derived at 50% of input.
OpenAI — GPT-5 Pro batchOutput Derived at 50% of output.
OpenAI — GPT-5 Nano cachedInput Derived at 10% of input.
OpenAI — GPT-5 Nano batchInput Derived at 50% of input.
OpenAI — GPT-5 Nano batchOutput Derived at 50% of output.
Google — Gemini 3 Flash cachedInput Derived at 10% of input — Google caching discount convention ~90%.
Google — Gemini 3.1 Flash-Lite cachedInput Derived at 10% of input — Google caching convention.
Google — Gemini 3.1 Flash-Lite batchInput Derived at 50% of input — Google Batch API uniform 50% discount.
Google — Gemini 3.1 Flash-Lite batchOutput Derived at 50% of output — Google Batch API uniform 50% discount.
Google — Gemini 2.5 Pro cachedInput Derived at 10% of input.
Google — Gemini 2.5 Flash cachedInput Derived at 10% of input.
Google — Gemini 2.5 Flash-Lite cachedInput Derived at 10% of input — Google caching convention.
Google — Gemini 2.5 Flash-Lite batchInput Derived at 50% of input — Google Batch API uniform 50% discount.
Google — Gemini 2.5 Flash-Lite batchOutput Derived at 50% of output — Google Batch API uniform 50% discount.
xAI — Grok 4 (legacy) cachedInput Extrapolated at 25% of base.

Pricing is cross-verified against the LiteLLM community registry when available. Daily snapshots are kept in aicost_pricing_snapshots; every change is logged to aicost_price_changelog with old & new values for full audit trail. Read the full methodology →