# The Account AI-Review Gate Read: Council, Security-Led, Ad Hoc, or Open

*Grade any target account's AI approval gate from six public signals and leave with a gate type, named seats, a review-time band, and a pre-wire move before first touch.*

- Canonical URL: https://www.refolk.ai/guides/account-ai-review-gate-read
- Pillar: Sales and go-to-market
- Format: Framework
- Published: 2026-11-22
- Last reviewed: 2026-11-22
- Reading time: 23 min
- Keywords: how to find AI governance committee at target account, who approves AI vendors in enterprise, AI vendor risk review sales cycle length, map AI buying committee before outreach, find chief AI officer or AI council members at a company, account research AI governance signals

## Key takeaways

- Headline proxy figures overstate AI gates: in 2024 more than 31% of the S&P 500 disclosed some AI oversight, but only 11% disclosed explicit board or committee oversight.
- A verified enterprise-scope ISO/IEC 42001 certificate is a strong gate signal because only 271 existed worldwide at the end of 2025, against 125,622 ISO 27001 certificates.
- In Refolk's index there are 703 current US Chief AI Officers but only 73 current AI-governance title holders, so a CAIO title alone does not prove an AI council exists.
- Supplier terms that require approval before a vendor uses AI tools imply an intake gate; terms that require disclosure only upon request do not.
- Vendor-reported AI reviews run 8 to 12 weeks for high or regulated tiers but 5 to 10 business days for vendors holding SOC 2 Type II, ISO 27001 and documented AI governance, so your own evidence pack sets much of the cycle.
- Score six dimensions 0 to 2 on quotable evidence only, assign the gate type with a first-match rule, and have a second rep grade blind before the verdict counts.

This is the outside-in read I run before spending first touches on a target account for an AI product. It tells you which AI approval gate the account runs, who staffs it, and whether to pre-wire it, plan around it, or move the account to next quarter. It is written for founders selling their own product, account executives, SDR leads, and partnerships teams. You get six scored evidence dimensions, a first-match rule that turns the scores into one of four gate types, and a seat list, review-time band, and pre-wire action for each type. It is tight enough that two reps grading the same account reach the same verdict.

## What is an AI-review gate, and what are the four types?

An AI-review gate is the approval step between "the champion wants this" and "we can sign", where someone other than the buyer checks how a vendor uses AI. From public evidence, an account's gate falls into one of four types: Council, Security-led, Ad hoc, or Open.

The usual advice is to ask your champion who reviews AI purchases. That works on call two, but by then you have already chosen who gets the first touch and how to forecast the deal. The stall this read prevents is a council or security veto nobody saw until after the champion said yes. Grading the gate before contact turns that surprise into an account attribute you can plan around.

| Gate type | What the public evidence shows | Who staffs it | Default action |
|---|---|---|---|
| Council | Named management body or enterprise-scope certificate | Governance lead, report signer, security reviewer | Pre-wire: governance seat in first wave |
| Security-led | Approval verb in supplier terms, mandatory exposure, or board oversight only | Third-party risk, infosec, privacy | Plan around: champion requests vendor pack in meeting one |
| Ad hoc | A policy page or a posting, no standing body | Privacy, legal, or IT, whoever holds it | Plan around: arm the champion, map three reviewers |
| Open | Nothing on any dimension | Champion and standard procurement | Proceed, re-grade next quarter |

"Deprioritize" is not a fifth gate type. It is an action that can land on a Council or Security-led account when your own evidence pack is not ready and the review band runs past the weeks left in the quarter. The verdict section gives the exact rule.

## How much public evidence of AI oversight actually exists?

Less than the headline numbers suggest. ISS-Corporate found more than 31% of the S&P 500 disclosed some board oversight of AI in 2024, but only 11% gave specific disclosure of full board or committee oversight. That roughly 3x gap is mostly director expertise being counted as oversight.

ISS counts three things as "oversight": a board committee with oversight responsibility, a director knowledgeable about the technology, or an AI ethics panel. Only the first tells you anything about who might review a vendor. Read every headline figure with that in mind.

| Cohort | Year | Metric | Value |
|---|---|---|---|
| S&P 500 | 2022-23 | Any AI oversight | 15%+ |
| S&P 500 | 2024 | Any AI oversight | 31%+ |
| S&P 500 | 2024 | Explicit board or committee oversight | 11% |
| S&P 100 | 2025 | Board-level oversight | 54% |
| S&P 100 | 2025 | Oversight plus formal AI policy | 28% |

Sources are ISS-Corporate (rows 1 to 3), the Glass Lewis S&P 100 review (row 4), and a secondary summary (row 5). Definitions differ between sources, so treat the rows as direction, not as a time series.

#### S&P 100 AI oversight disclosure, 2025 proxies

| Stage | Figure | Note |
| --- | --- | --- |
| S&P 100 companies | 100% | All proxy statements reviewed |
| Board-level AI oversight disclosed | 54% | Full board or a committee |
| Oversight plus a formal AI policy | 28% | Secondary summary of the same cohort |

*Even among the largest US companies, only 28% disclosed both board oversight of AI and a formal AI policy.*

A few more facts change how you read a filing:

- **Committee versus full board.** Of the S&P 100 companies disclosing board oversight, a secondary summary puts 63% with a specific committee, most often audit or technology, and 37% with the full board.
- **Committees are growing.** EY's 2025 proxy review found the number of S&P 500 companies naming a committee with AI oversight more than tripled, with audit committees the primary choice.
- **Expertise is not a gate.** Nearly half of Fortune 100 proxies cited AI in director qualifications in 2025, almost double the 26% in 2024. That is the easiest thing to misread as oversight.
- **Sector matters.** In the 2022-23 cycle, IT companies disclosed AI oversight at 38% and Health Care at 18%. A later sector breakdown is not established publicly.

Certification is rarer still. ISO/IEC 42001 is the international standard for an AI management system, meaning the policies and controls an organisation uses to govern its own AI. The ISO Survey counted 271 ISO/IEC 42001 certificates worldwide at the end of 2025, against 125,622 for ISO 27001.

**1 in 464 - ISO/IEC 42001 certificates per ISO 27001 certificate worldwide, end of 2025**

271 against 125,622, which is why a verified enterprise-scope 42001 certificate is a strong marker of a formal AI gate.

The practical consequence is that most accounts, including large ones, score low on filings, and private companies file no proxy at all. A blank filing is weak evidence that there is no gate. That is why the read uses six dimensions rather than leaning on the proxy.

## How do you score the six evidence dimensions?

Score each dimension 0, 1, or 2, and only on evidence you can quote with its source and section. The total runs from 0 to 12 and tells you how much evidence you hold. The gate type comes from which dimensions scored, using the first-match rule in the next section.

| Dimension | 0 | 1 | 2 |
|---|---|---|---|
| D1 Board or committee oversight in filings | None, or director expertise only | Full board holds AI oversight | Named committee whose charter or description names AI |
| D2 AI policy or responsible-AI report | None | Principles page, no named owner or body | Report names a management body or signing executive |
| D3 ISO/IEC 42001 certification | None, or "aligned with" | Product scope, or accreditor unverified | Enterprise scope, certifier accredited for 42001 |
| D4 Governance seats | None | Open posting only | Current holder of a governance title |
| D5 Sector and jurisdiction exposure | None | Regulated sector, but your use is out of scope or not yet applicable | Your use falls in a mandatory regime now in effect |
| D6 AI terms in supplier pages | None | Disclosure duty ("inform" or "upon request") | Approval required before vendor AI use |

> **Rule:** Quote it or score it zero
>
> A dimension scores above zero only when the gate card holds a quoted phrase, its source, and its section. Impressions, press coverage, and "they seem serious about AI" score zero, which is what lets two reps reach the same verdict.

The dimensions sit at different depths of the organisation. Evidence at an outer layer does not prove the inner layers exist.

#### Where each dimension sits in the gate

1. **External mandate** - Regulation that forces a review regardless of disclosure (D5)
2. **Board** - Committee or full-board AI oversight in the proxy (D1)
3. **Management body** - Responsible-AI report naming a council, and 42001 scope (D2, D3)
4. **Named seats** - Current governance title holders and open postings (D4)
5. **Intake desk** - Supplier terms and the security questionnaire a vendor actually meets (D6)

*A board committee that watches AI risk is not proof of a management body or an intake desk beneath it.*

### D1: Board or committee oversight in filings

**What it proves:** the board has assigned someone to watch AI risk. Read the risk-oversight section of the latest DEF 14A, the annual proxy statement US public companies file, along with the committee charters. Delegated-committee language looks like Citi's Technology Committee, which "reviews trends... including... Artificial Intelligence and Machine Learning."

**When it lies:** the only AI mention is in director bios, or a board committee exists with no management body beneath it.

### D2: Published AI policy or responsible-AI report

**What it proves:** someone in management owns AI conduct in public. Record who signs it and whether it names a council, committee, or office.

**When it lies:** it is a principles page with no owner. That scores 1, not 2.

### D3: ISO/IEC 42001 certification with verified scope

**What it proves:** when the certificate is enterprise-scoped and the certifier is accredited, a formal AI management system exists and has been audited. Verify in two passes:

1. Confirm the certification body is accredited for 42001 itself, on the accreditation body's own register (ANAB, UKAS, RvA, JAS-ANZ, SAC or IAS) or on IAF CertSearch. Accreditation for ISO 27001 does not extend to 42001.
2. Read the certificate's scope statement and class it as enterprise or product.

**When it lies:** the certificate covers a product line. AWS's initial November 2024 certificate named Bedrock, Q Business, Textract and Transcribe. That shows those services are governed, not that procurement runs an AI gate. "Aligned with 42001" is a self-declaration and scores 0. How often certificates cover only a product line is not established publicly.

### D4: Governance hiring and named roles

**What it proves:** there are people whose job is the gate. Current holders count for more than open postings. One posting analysis found around 71 new US AI governance postings a week, 56% of them from companies with 10,000+ employees.

**When it lies:** a single req at a mid-size firm is a plan, not a standing body. Growth estimates also conflict: IAPP, citing Stanford HAI, reports 17% year-over-year growth, while another source cites 150%. Do not use hiring momentum as a score input.

### D5: Regulated-sector and jurisdiction exposure

**What it proves:** an outside rule may force a gate regardless of what the account publishes. Check HQ and EU operations, the sector (banking over $30 billion in assets, insurance, health, HR tech), and whether your product falls in an EU Annex III high-risk use or a consequential-decision use.

**When it lies:** the dates or scope in the rep's head are stale. These are the dates as published. Re-check each one against the regulation text or a current legal summary before you score it or cite it to a buyer.

| Regime | Date as published | What it means for the gate |
|---|---|---|
| EU AI Act, Article 50 transparency | Most obligations apply 2 Aug 2026 | Transparency duties, not a high-risk gate |
| EU Digital Omnibus, Regulation (EU) 2026/1744 | In force 27 Jul 2026; Annex III 2 Dec 2027; Annex I 2 Aug 2028 | High-risk duties postponed |
| Colorado revised AI law | Effective 1 Jan 2027 | Consequential-decision uses |
| California ADMT regulations | Enforcement from 1 Jan 2027 | Automated decision-making uses |
| US banks, SR 26-2 / OCC Bulletin 2026-13 | Issued 17 Apr 2026, replacing SR 11-7 | Generative and agentic AI left out of formal scope |

Two details catch people. Generative AI already on the EU market before 2 Aug 2026 has until 2 Dec 2026 for machine-readable marking. One source claims Article 50 applied from 2025; that is an outlier, so do not use it.

### D6: AI-vendor terms in procurement or supplier pages

**What it proves:** the account has written an intake rule for vendors that use AI. Record the verb, because the verb is the score:

- **Approval, scores 2.** OpenAI's supplier code says suppliers "shall disclose and obtain approval for AI tools, model providers, or subprocessors."
- **Affirmative disclosure, scores 1.** UPS's AI Addendum says the vendor "shall inform UPS in writing about the usage of any AI Technology."
- **Disclosure on request, scores 1.** OpenText's supplier code requires disclosure "upon request", the weakest form of intake.

**When it lies:** disclosure language gets read as an approval gate. None of these documents, in the excerpts I have seen, names the function that owns AI intake. D6 tells you a gate exists, not who staffs it.

## Who staffs the gate, and how do you build the seat list?

Start the seat list with current holders of AI-governance titles, add the executive who signs the responsible-AI report, and add a Chief AI Officer only when a governance seat reports to them. These seats are rare, so finding even one named person changes the first-touch plan.

| Market | AI-governance titles | "Head of AI Governance" in sample of 25 | "Responsible AI Lead" in sample of 25 |
|---|---|---|---|
| US | 73 | 2 | 2 |
| UK | 43 | 4 | 5 |

In Refolk's index of professional profiles, there are 73 current holders of AI-governance titles in the US and 43 in the UK, a UK-to-US ratio of 0.59. Those counts use exact-title match. That matters, because the title is unstable: "AI Governance Lead" competes with AI Compliance Officer, Responsible AI Lead, Head of AI Ethics, and AI Risk Manager. Search all five.

**9.6 to 1 - US Chief AI Officers per current AI-governance title holder, in Refolk's index**

703 current US Chief AI Officer title holders against 73 AI-governance title holders, so the CAIO title is far more common than the seat that actually runs a gate.

Where the seats sit differs by market. In the UK sample of 25 from Refolk's index, the top employers include NatWest Group (two profiles), Legal & General, AXA UK and Howden, and 4 of the top 10 employers are banks or insurers. The US sample mixes public sector and consultancies, including JPMorganChase, the Defense Intelligence Agency and Maryland DoIT. The UK pattern is consistent with regulator pressure on financial firms, but that is my inference and it is unverified.

Use these seat lists by gate type:

- **Council:** the governance lead under any of the five titles; the signer of the responsible-AI report; the security or third-party risk reviewer who runs the questionnaire; the CAIO only if a governance seat reports to them.
- **Security-led:** the third-party or vendor risk manager; the information security reviewer; privacy. For a generative AI product at a US bank over $30 billion in assets, SR 26-2 leaves generative and agentic AI out of formal model-risk scope. One practitioner summary says those systems fall back on enterprise risk, information security, privacy and vendor controls. Plan on vendor risk and security as reviewers, not model validation, but confirm locally.
- **Ad hoc:** IAPP data splits AI governance ownership across privacy (22%), legal (22%) and IT (17%). Name one person in each until the champion tells you who holds it.
- **Open:** the champion and standard procurement.

Building this list by hand means running five title variants against each account, then checking whether each match is current and at the right company. [Refolk](/) takes the seat as a plain-English request and returns current holders across LinkedIn and the open web, so the variant problem is handled in the query rather than in your spreadsheet.

I ran this search: `Heads of AI governance or responsible AI leads at UK banks and insurers with over 5,000 employees` - [see the full result list](https://www.refolk.ai/s/bd4gs6ebwn).

*Returns current holders of governance seats at large UK financial firms, with title, employer, and profile links you can drop straight into a gate card seat list.*

## How do you turn scores into a gate type, a review band, and an action?

Apply the first-match rule: work down the table, and the first row whose condition is true sets the gate type. The gate type then sets the review band and the pre-wire action. Adjust the band for your own attestations, because much of the delay sits with the vendor, not the gate.

| Order | If this is true | Gate type | Review band to forecast |
|---|---|---|---|
| 1 | D2 = 2, or D3 = 2, or D4 = 2 with D2 = 1 | Council | 8 to 12 weeks |
| 2 | D6 = 2, or D5 = 2, or D4 = 2, or D1 is 1 or 2 | Security-led | 3 to 6 weeks; 8 to 12 if D5 = 2 |
| 3 | Any other non-zero score | Ad hoc | 4 to 8 weeks |
| 4 | Every dimension at 0 | Open | No AI-specific leg established |

The bands are my mapping of vendor-reported durations onto gate types. Here are the source figures:

| Condition | Duration | Source type |
|---|---|---|
| Standard risk tier | 3 to 6 weeks | Vendor blog, methodology unpublished |
| High or regulated tier | 8 to 12 weeks | Vendor blog, methodology unpublished |
| Vendor with no attestations | 4 to 8 weeks | Vendor blog, methodology unpublished |
| SOC 2 Type II + ISO 27001 + documented AI governance | 5 to 10 business days | Vendor blog, methodology unpublished |

The AI-specific block is described as typically 30 to 60 items inside a broader enterprise security review. CSA's AI-CAIQ, a standardised AI questionnaire, runs to 320 questions, according to a vendor blog. How many extra reviewers an AI review adds is not established publicly.

**Attestation modifier.** If you hold SOC 2 Type II, ISO 27001 and documented AI governance, shorten Security-led and Ad hoc bands toward 5 to 10 business days. I would not shorten a Council band below 3 to 6 weeks until your own closed deals show it.

**Pre-wire actions by gate type:**

- **Council: pre-wire.** Put the governance seat in the first wave alongside the champion. Send the AI governance pack before anyone asks: model providers, subprocessors, attestations, and pre-filled answers to the AI questionnaire block.
- **Security-led: plan around.** Have the champion request your vendor pack in meeting one rather than waiting for the questionnaire to arrive.
- **Ad hoc: plan around.** Give the champion the pack plus a one-page note written for privacy, legal and IT, since any of them may end up holding the review.
- **Open: proceed.** Run the normal motion and re-grade next quarter.

#### Gate formality against your pack readiness

Horizontal axis runs from Informal gate (Open, Ad hoc) to Formal gate (Security-led, Council). Vertical axis runs from Pack not ready to Pack ready (attestations plus AI governance docs).

| Quadrant | What it means |
| --- | --- |
| Ready, informal gate | Move fast; ship the pack with the first meeting |
| Ready, formal gate | Pre-wire the governance and security seats now |
| Not ready, informal gate | Plan around; arm the champion while you build the pack |
| Not ready, formal gate | Deprioritize this quarter unless the band fits |

*A formal gate is only a reason to wait when your own evidence pack is not ready for it.*

> **Rule:** When to deprioritize this quarter
>
> Move an account to next quarter when the gate is Council, or Security-led with D5 = 2, your attestation set is incomplete, and the top of the review band runs past the weeks left in the quarter. Keep it on next quarter's list with the gate card attached; do not drop it.

## What is the step-by-step procedure for grading an account?

Run the eight steps below in order, about 60 minutes of research without a certificate to verify and about 95 with one, then score and calibrate. Each step ends with a dimension scored on the gate card.

#### Grade one account's AI-review gate

1. **Screen regulatory and sector exposure** - Check HQ, EU operations, sector, and whether your product falls in an Annex III or consequential-decision use, re-checking the regime dates. About 10 minutes; done when D5 is scored.
2. **Read the supplier terms for an AI intake verb** - Find an AI addendum or supplier code and quote the verb: approval, inform in writing, or upon request. About 10 minutes; done when D6 is scored.
3. **Read the proxy and committee charters** - Classify the AI disclosure in the latest DEF 14A as committee, full board, expertise-only, or none. 15 to 20 minutes; done when D1 has a quoted phrase and section.
4. **Check for an AI policy or responsible-AI report** - Note who signs it and whether it names a management body. About 10 minutes; done when D2 is scored.
5. **Verify any ISO/IEC 42001 certificate and its scope** - Confirm the certifier's 42001 accreditation on the accreditor's register, then read the scope statement. About 30 minutes; done when scope is classed enterprise or product.
6. **Scan for governance seats and postings** - Search current holders and open postings across all five title variants, plus any CAIO with a governance seat reporting in. About 15 minutes; done when D4 is scored and seats are named.
7. **Score, assign the gate, and set the action** - Apply the first-match rule, then fill the seat list, review band with attestation modifier, and pre-wire action. Done when the gate card is complete.
8. **Calibrate with a blind second grade** - A second rep grades without seeing the first card, and both reconcile dimension by dimension on the quotes. Done when there is one agreed verdict.

**A note on order.** The procedure as first assembled put the supplier-terms check sixth. Practitioner guides on security review argue for security evidence first, with the champion requesting the vendor pack rather than waiting for the questionnaire. I moved supplier terms to step 2 for that reason: an approval verb tells you, before you open a proxy, that a vendor pack has to ship with the first meeting.

**AI-review gate card**

```
Account:
Graded by:                Second grader (blind):
Date graded:

D1 Board or committee oversight (0/1/2):
   Quote + proxy section:
D2 AI policy or responsible-AI report (0/1/2):
   Quote + signer + named body:
D3 ISO/IEC 42001 (0/1/2):
   Accreditor register checked:      Scope statement:
D4 Governance seats (0/1/2):
   Named holders:                    Open reqs:
D5 Exposure (0/1/2):
   Regime:                           Date re-checked on:
D6 Supplier AI terms (0/1/2):
   Intake verb (approval / inform / upon request) + source:

Total (0-12):
First-match row fired: 1 / 2 / 3 / 4
Gate type: Council / Security-led / Ad hoc / Open
Seat list:
Review band forecast:            Attestation modifier applied: yes / no
Action: Pre-wire / Plan around / Proceed / Deprioritize this quarter
Re-grade trigger:
```

*One card per account in the CRM account record. Every non-zero score needs a quote and a source line.*

## How does the gate read go wrong?

Most errors come from reading a signal at the wrong layer: board for management, product for enterprise, disclosure for approval. Each failure mode below says what happens and the check that catches it.

1. **Proxy "oversight" is really expertise only.** ISS counts AI-skilled directors as oversight, so headline figures include accounts with no assigned body. *Check:* require a charter or committee-description sentence that names AI. Director bios alone score zero.
2. **A committee exists, but only at board level.** An audit committee overseeing AI risk is not evidence of a management-level intake council. *Check:* look for a management body named in the responsible-AI report. Without one, the account cannot reach Council through D1.
3. **The certificate is scoped to a product.** AWS's first certificate named four services, not procurement. *Check:* read the scope statement. "Aligned with 42001" is a self-declaration and scores zero.
4. **A false negative from the register.** IAF ceased operations on 1 January 2026, CertSearch still runs, and one directory reports it may not show ISO/IEC 42001 in accredited scopes. A missing entry proves nothing. *Check:* go to the accreditor directly, such as ANAB or UKAS.
5. **The CAIO title is noise.** The 703 US Chief AI Officer profiles in Refolk's index include small firms and mis-titled CEO and CTO profiles. *Check:* confirm company size and whether a governance seat reports to the CAIO before adding them to the seat list.
6. **The posting is aspirational.** A single "AI Governance Manager" req at a 500-person firm is not a standing body. *Check:* look for current holders plus a policy page. How long a posting takes to become a disclosed committee is not established publicly, so do not forecast one.
7. **Stale regulatory read.** Reps who cite SR 11-7, or who carry the original August 2026 EU high-risk dates, will mis-forecast regulated accounts. *Check:* re-check every D5 date against a current source and record the re-check date on the card.
8. **Supplier code mistaken for intake.** OpenText requires disclosure "upon request" while OpenAI requires "approval". *Check:* only approval language scores 2 and implies a gate.

> **Watch out:** The most expensive misread is a board committee scored as a council
>
> It sends the first touch to nobody in particular and the forecast to the 8 to 12 week band for a gate that may not exist. Then the real reviewer, usually vendor risk, turns up late anyway.

> A board that watches AI risk is not a desk that approves AI vendors.

There is also a structural limit. Every review duration here is vendor-reported with unpublished methodology, and no public source documents this procedure end to end. The scoring rule is a standard for consistency, not a validated predictor. Treat the bands as priors until your own data replaces them.

## How do you keep the read calibrated and current?

Calibrate by blind double-grading a sample of accounts each quarter, and keep the read current by re-grading on specific triggers rather than on a calendar alone. Over time, replace the vendor-reported bands with review durations measured on your own deals.

**Calibration.** When two reps disagree on the gate type, the cause is almost always one dimension. Reconcile on the quotes. If a quote does not settle it, score the lower value and log the open dimension as the first question for the champion. That keeps the verdict conservative and turns the disagreement into discovery.

**Re-grade triggers:**

- A new proxy statement or committee charter is filed.
- A responsible-AI report is published or re-signed.
- A governance title holder joins, leaves, or a new req appears.
- A D5 regime date passes or is amended.
- The supplier code or AI addendum is revised and the intake verb changes.

Of these, the seat trigger is the hardest to watch by hand, because governance titles move between five variants and people change roles without notice. Re-running the same plain-English seat query in Refolk each quarter and diffing it against the seat list on the card is the cheapest way to catch it.

**Replacing the bands.** For every closed or stalled deal, record the gate type you assigned, the date the AI or security review started, and the date it cleared. After a handful of deals per gate type, your own medians are better than any figure in this guide.

#### Before the verdict counts

- [ ] Every non-zero dimension on the gate card has a quoted phrase, a source, and a section.
- [ ] D1 is scored from a charter or committee description, not from director bios.
- [ ] Any 42001 certificate was checked against the accreditor's own register and its scope classed as enterprise or product.
- [ ] D5 dates were re-checked against a current source and the re-check date is on the card.
- [ ] D6 records the exact intake verb, and only approval language scored 2.
- [ ] The seat list names people under all five governance title variants searched, and any CAIO listed has a governance seat reporting to them.
- [ ] The review band reflects the attestation modifier for your current evidence pack.
- [ ] A second rep graded the account blind and any unresolved dimension is logged as a champion question.

## Frequently asked questions

### How do I find the AI governance committee at a target account?

Start with the proxy statement's risk-oversight section and the committee charters, then look for a responsible-AI report that names a management body. A board committee that oversees AI risk is not the same as a management council that reviews vendors, so you need the second document to confirm an intake body. For private companies with no proxy, rely on the responsible-AI page, current governance title holders, and the supplier terms.

### Who approves AI vendors in an enterprise?

It depends on the gate type. Where a standing council exists, an AI governance or responsible-AI lead usually sits on it alongside a security reviewer. Where there is no council, AI questions are folded into the security or third-party risk review. Where ownership is ad hoc, IAPP data splits AI governance ownership across privacy (22%), legal (22%) and IT (17%), so map all three until you know which one holds it.

### How long does an AI vendor risk review add to the sales cycle?

Vendor-reported figures put a standard-tier review at 3 to 6 weeks and a high or regulated tier at 8 to 12 weeks, falling to 5 to 10 business days for vendors holding SOC 2 Type II, ISO 27001 and documented AI governance. None of these sources publishes its methodology, so use them as starting bands and replace them with durations measured on your own closed deals.

### Does a Chief AI Officer at the account mean there is an AI council?

No. Refolk's index holds 703 current US Chief AI Officer title holders against 73 current AI-governance title holders, about 9.6 to 1, and the CAIO pool includes small firms and mis-titled CEO or CTO profiles. Count a CAIO toward the gate only when the company is large enough to run one and a governance seat demonstrably reports to them.

### Is an ISO 42001 certificate proof the buyer runs a formal AI gate?

Only when the certifier is accredited for ISO/IEC 42001 specifically and the certificate's scope covers the enterprise rather than a product line. Accreditation for ISO 27001 does not carry over, and "aligned with 42001" is a self-declaration. Because only 271 certificates existed worldwide at the end of 2025, a verified enterprise-scope certificate is a strong signal of a formal gate.

### Why not just ask the champion who reviews AI purchases?

Ask anyway, but grade first. By the time you can ask, you have already chosen who gets the first touch and how to forecast the deal. The stall this read prevents is a council or security veto that surfaces after the champion says yes. A pre-contact grade lets you put the governance seat in the first wave and ship the vendor pack with meeting one.

---

*From the Refolk guide library. I revise these guides rather than replacing them, so the current version is always at https://www.refolk.ai/guides/account-ai-review-gate-read*
