AI governance quiz - Pacific AI

The AI Governance Quiz: Strategy or Prayer?

Most organizations think they're governing AI. They're not - they're crossing their fingers 🤞

Take this 10-question test to find out if your governance is operational (strategy) or wishful thinking (prayer).

Time: 3 minutes

Questions: 10 (4 points each, 40 total)

Result: Instant score + personalized roadmap

Between the California TFAIA, Texas RAIGA, and new Federal Executive Orders, the legal goalposts are moving daily. How do you stay ahead?

The "Ostrich Strategy." We bury our heads in the sand and wait for the cease-and-desist.

A "Living Document" (a SharePoint spreadsheet that died in 2023).

We pay a law firm $800/hour to send us monthly PDFs that are outdated by the time we finish our coffee.

Automated Policy Mapping. Our system de-duplicates global and state rules into a single, self-updating set of controls.

None

A team wants to launch a new Gen AI tool. What is the current "Approval Experience"?

Total chaos. The product team ships the Gen AI tool, and the security team can only hope it doesn't expose payroll data.

The "Black Hole." The product team submits a terrible Word doc and the governance team lets it disappear into the void for 8 weeks with no response.

The "Interrogation." The governance team traumatizes the engineers with absurd 60-page forms full of zero-value compliance questions.

AI-Augmented Triage. Our system auto-ingests project docs to generate a high-quality risk draft in minutes, focusing only on real, actionable threats.

None

How many AI models or third-party wrappers are running in your company right now?

Does "Kevin from Marketing using ChatGPT" count? Because we have no clue.

It's a scavenger hunt. Give me three days and 40 Slack messages and I’ll give you a list.

We have a registry, but it's a static graveyard that teams only update when someone nudges them.

We have a living registry. A single source of truth for every project, vendor, and risk level, connected directly to our development pipelines.

None

When is the "Governance" part of an AI project considered "done"?

When the lawyer stops emailing us.

After we pass the initial "Go/No-Go" gate before the first deployment.

We do an annual review, which is basically an archaeology expedition into last year’s code.

We have a continuous governance. Risk scores update automatically in a living loop whenever the model, data, or regulations change.

None

How do you ensure your LLM isn't confidently lying to your customers or being accidentally biased?

The "Vibe Check." We asked it three questions, it didn't swear at us, so we went live.

"Red Teaming" (which is just our intern trying to make the bot say something offensive).

We use generic open-source benchmarks that have nothing to do with our actual industry (Healthcare/Finance).

We run automated, industry-specific tests (like MedHELM for healthcare) before every release.

None

How do your AI engineers get their models approved by the Compliance team?

They don't – compliance finds out about new AI when it's already in production.

Engineers avoid compliance until the last second, then scramble to fill out paperwork nobody reads.

They manually copy-paste test results into our internal portal, like it's 1999.

Test results automatically flow from their CI/CD pipeline – if safety tests don't pass, the build is blocked.

None

The model is live. How do you know if its accuracy has dropped or if bias has crept in?

We wait for someone to screenshot our bot being weird and post it on social media.

We check if the server is still "on." Beyond that, it’s a black box.

A data scientist manually audits a random sample of logs once a month.

We have automated alerts that notify us the instant accuracy or fairness metrics degrade.

None

You've built an AI Agent that can actually take actions (booking, prescribing, etc.). How do you test its "logic"?

We just cross our fingers and hope it doesn't delete the database or spend $50k on paperclips.

We try to imagine every possible mistake Gen AI it could make (spoiler: we can't).

We keep a human in the loop for every single click, which makes the agent uselessly slow.

We simulate thousands of scenarios to catch emergent problems before the agent reaches real users.

None

What happens when your Compliance team sits down to evaluate an engineer's technical evidence?

Pure confusion – it's like a conversation between a dolphin and a helicopter.

"Hostage Negotiations." We have 4-hour meetings where everyone leaves more confused than they started.

Compliance has a "cheat sheet," but they still can't tell if the engineer is explaining a model or reciting a sourdough recipe.

Technical data is auto-translated into plain language with clear Red/Yellow/Green risk ratings.

None

10.

A regulator walks into your office and asks for a full audit trail and a "Model Card." What happens?

I start updating my resume and LinkedIn profile.

Can you give us a month? We need to find the person who wrote the original prompt.

We spend the weekend frantically cobbling together screenshots and emails into a "report".

One-click export – our system generates versioned, audit-ready Model Cards from real-time data.

None

Time's up

AI Risk Management Framework

ISO/IEC 42001:2023 AI management systems

Draft Guidance for Developers of AI-Enabled Medical Devices

AI Risk Management Framework

ISO/IEC 23894:2023 Guidance on AI risk management

Guidance on AI in Employment Law

AI Risk Management Framework

5259-4:2024 Data quality for analytics and machine learning

Assurance Standards Guide

AI Risk Management Framework

ISO/IEC NP TS 12831 Testing for AI Systems

AI Risk Management Framework – Generative AI Profile

Privacy Framework (PF)

Cybersecurity Framework (CSF)