All press releases

For Immediate Release

vCISO Lite Launches the AI Sub-Processor Register — a Citation-Anchored Record of Which AI Providers Sit Under Your Vendor Stack, What They Do With Your Data, and What Nobody Has Verified Yet

Discovery reads what vendors have already published and extracts their AI posture with a citation on every claim; nothing becomes the record until a person on the customer's team confirms it, and anything uncited stays an honest "unknown." Six posture dimensions across the whole vendor list, per-dimension concentration made visible, judgements that name the rule and framework they derive from, and a one-click PDF for the auditor, the board, or the customer security review. Included in vendor risk on all vCISO Lite plans, live today.

ATLANTA — September 22, 2026 — vCISO Lite today launched the AI Sub-Processor Register, an org-wide record of what every vendor’s AI does with a customer’s data: which model providers sit underneath the vendor stack, whether customer data trains anyone’s models, and what nobody has verified yet. Every discovered claim in the register carries a citation back to the vendor’s own published disclosures, nothing becomes the record until a person confirms it, and anything that cannot be cited is rendered as unknown rather than guessed.

The register answers a question buyers are being asked with increasing frequency and can rarely answer: which AI providers actually process my data through my vendors, and how much of my stack quietly depends on the same one? Industry research puts numbers on the gap. Venminder’s 2025 State of Third-Party Risk Management report found 23% of organizations still don’t monitor vendor AI use at all, and Ncontracts’ 2026 survey of financial institutions found AI has risen for the first time to tie cybersecurity as their top third-party risk concern. Most tools respond with either an engineer’s model inventory or a one-time questionnaire that rots the day it’s filed. vCISO Lite built the third thing: a standing register, held to the evidentiary standard of the vendor-risk program it lives inside.

“Ask a company which AI providers touch their data through their vendors and you get a shrug, a stale spreadsheet, or a guess dressed up as an answer. The guess is the worst of the three, because you can’t audit it,” said Yolonda Smith, founder of vCISO Lite. “We built the register to be honest about the difference. Every claim in it points at the sentence it came from, a person on your team signs off before it counts, and where we couldn’t verify something the register says unknown: out loud, on the screen, in the export. An unknown you can see is something you can go close. A guess is something you defend in front of an auditor and lose.”

What’s in the release

The AI Sub-Processor Register is live today for every vCISO Lite customer, inside the vendor-risk surface they already run. In this release, customers can:

  • Discover vendor AI posture with receipts. Discovery reads what each vendor has already published (sub-processor pages, DPAs, trust centers, privacy and AI disclosures) and extracts the posture: which models and providers they use, for what purpose, under what commitments. Every extracted field carries a citation with the source URL and the quoted passage it came from. A field with no citation stays unknown; discovery never fills a blank with a guess.
  • Confirm before anything counts. Discovered entries land marked for review and stay that way until someone on the customer’s team confirms or corrects them on the vendor’s AI Posture tab. Confirming stamps who attested and when, so the register is the organization’s attested position, not a scrape.
  • See the whole stack on one screen. The org-wide register covers six dimensions per vendor: data flows, consent basis, training data, human authority, auditability, and incident response. The dimension headers double as filters, so “show me every vendor we haven’t confirmed training-data terms for” is one click.
  • See concentration before it surprises them. Per-dimension bars show how much of the vendor stack resolves to the same underlying posture or provider. Four vendors with four logos can be one model provider wearing four coats; the register is where that becomes visible. The composition question is a standing theme of the company’s vendor-risk work; see “Vendor Concentration Risk: The Dimension Per-Vendor TPRM Misses”.
  • Read judgements that show their work. Where the register evaluates a vendor’s posture against an expectation, the cell names the rule it applied and the framework the rule derives from, and flags rules that are past their review date instead of presenting stale judgement as fresh.
  • Fold in what vendors already told them. Evidence vendors have provided in security questionnaires feeds the same picture, so the register reflects what a vendor told this customer, not only what it published to the world.
  • Export it for the people who ask. One click produces a PDF of the complete register, every vendor with the basis for each judgement, for the auditor, the board, or the customer security review that asked what AI touches their data.

A register an auditor can actually use

The design position behind the register is that provenance is not a feature of an AI vendor record — it is the record. A register that silently fills gaps with plausible answers fails at the exact moment it exists for: the audit, the diligence request, the customer security review where someone asks “how do you know?” So the register enforces three properties end to end: every discovered claim is citation-anchored to a source a reviewer can open, every confirmed claim names the person who attested it and when, and every gap is rendered as unknown rather than papered over. What the register could not verify is presented as plainly as what it could.

Not an AI bill of materials

The register arrives alongside a wave of AI bill-of-materials (AIBOM) work: component inventories, in formats like CycloneDX’s ML-BOM, that let a software producer enumerate the models, datasets, and frameworks inside a product. That work is valuable, and it is not what the register is. An AIBOM is the vendor’s inventory of what is inside their software: produced from the vendor’s vantage point, one product at a time, and only as available as each vendor chooses to make it. The register is the customer’s attested record of what every vendor’s AI does with the customer’s data. It is assembled from the buying side, across the whole vendor stack at once, and judged against the customer’s own regulatory reality.

The practical difference shows up in the questions each can answer. An AIBOM can tell you a product embeds a particular model. It cannot tell you whether your records are excluded from that model’s training, under which contractual commitment, who on your team verified it, and when, or that nobody has. Those are the register’s questions, asked identically of every vendor whether or not they publish anything machine-readable, with a citation where the answer exists and an honest unknown where it doesn’t. An inventory tells you what is there. The register tells you where you stand.

The two are complements, not rivals. As AIBOMs become a standard fixture of vendor disclosure, each one is simply more published material for the register’s discovery to read and cite. The register remains the layer where what a vendor discloses becomes what a customer has confirmed.

The pressure making this record necessary is regulatory as well as commercial. The EU AI Act places transparency and information-sharing duties across the AI supply chain, which means companies increasingly need to know, and be able to show, what sits underneath the AI features their vendors sell them; the company’s analysis of the Act’s record-keeping expectations is in the AI Governance series. At the same time, AI sub-processor disclosure is becoming a standard fixture of DPAs and trust centers, and enterprise security reviews now routinely ask vendors what AI touches customer data. Those questions are covered in the company’s guidance on evaluating AI in vendor products and across the vendor-risk editorial series. The register is built so those questions get answered from an attested record with citations, not reconstructed from memory under deadline.

Availability

The AI Sub-Processor Register is live today, included in vendor risk on all vCISO Lite plans, no separate SKU. Customers will find the org-wide register in their admin console and an AI Posture tab on every vendor assessment; see the vendor risk overview for where it fits, and the changelog entry for release detail.


About vCISO Lite

vCISO Lite is a compliance and cyber risk platform for growing companies that don’t have a full-time CISO. The platform helps customers close compliance gaps (SOC 2, ISO 27001, PCI DSS, HIPAA, and more), quantify cyber risk in the language their board and deal teams already speak, and, with the AI Sub-Processor Register, answer what AI actually touches their data across the vendor stack, with citations. vCISO Lite is headquartered in Atlanta, Georgia. Learn more at vcisolite.com, read related product releases in the changelog, or explore the vendor risk surface.

Media Contact

Press & Analyst Inquiries
Yolonda Smith, Founder
[email protected]

###