We run Vanta. We also run AiT Audit. Here is why both.

Vanta filed for IPO in March 2026. Drata closed a $200M Series C at a $4.5B valuation in February. The compliance automation category is now a real category, and the leaders earned it. We use Vanta on the Intelligent IT tenant. We recommend it to clients in regulated industries every week. None of what follows is a knock on Vanta, Drata, Secureframe, or Tugboat Logic.

It is, instead, a structural observation about what those platforms were designed to do and what they were not.

What Vanta and Drata are great at

The category solved the most expensive part of compliance: evidence collection. Before Vanta, a SOC 2 Type II audit meant six weeks of an internal owner chasing spreadsheets and screenshots. Vanta wires up to 200+ integrations (AWS, Okta, GitHub, Jamf, CrowdStrike) and pulls evidence continuously. Drata does the same thing with a different UX. Secureframe added MSP-friendly multi-tenant features in 2025.

For SOC 2, ISO 27001, HIPAA, and PCI DSS, this works. The auditor gets a clean export, the policies are versioned, the access reviews are timestamped. According to a Gartner report from January 2026, organizations using a compliance automation platform completed Type II audits 47% faster than those running manually. The numbers are real.

What they were not designed to do

Compliance automation platforms verify that controls exist. They do not verify that the code those controls protect actually does what the policy says it does.

That distinction matters more in 2026 than it did in 2022. Three trends pushed it from edge case to core gap.

AI/ML systems do not have controls Vanta can read

If your stack includes a Claude Sonnet endpoint, an OpenAI-powered chatbot, an embeddings index, or a model fine-tuning pipeline, Vanta has nothing to integrate with. The OWASP LLM Top 10 (2025 revision) and the NIST AI 600-1 Risk Management Framework define the controls. Vanta does not pull evidence on prompt-injection defenses, output filtering, or training-data provenance because there is no API to pull from. The control is in the code.

The 2025 SaaS supply-chain breaches changed what auditors look for

The Drift breach in August 2025, the Salesloft compromise in October, and the disclosed Slack token leakage from a third-party Workato integration in December all had the same shape: a legitimate OAuth token with broader scopes than the integration actually needed, sitting in a vendor's database, exfiltrated on compromise. Vanta will tell you the integration exists. Vanta will not tell you the integration's OAuth scope grant is wider than necessary, because Vanta is not reading your IAM bindings against your code's actual call patterns.

Continuous deployment changes the audit window

If your team ships 40 times a week, the gap between “control was in place at the moment of the screenshot” and “control is in place right now” is meaningful. Vanta's continuous monitoring is continuous against the integration surface. It is not continuous against the code surface.

The gap, concretely

Last quarter we ran AiT Audit against a mid-market SaaS company that had been Vanta-clean for two years. Type II reports immaculate, customer-facing trust portal green across the board. AiT Audit found:

• Three Cloud Run services where the runner service account had owner-tier IAM on the project, not least-privilege scoped.
• A Stripe webhook handler that did not verify the Stripe-Signature header on one of four endpoints. Anyone with the URL could replay a webhook.
• A Slack bot integration where the OAuth token had channels:read, files:read, and im:history when the actual code only used chat:write.
• A Supabase RLS policy that referenced auth.uid() instead of auth.uid()::text, which fails open under specific Clerk identity-token conditions documented in their February 2026 advisory.
• An OpenAI API key checked into a private GitHub Gist (not the main repo, but discoverable via gh search) from a 2024 debugging session that nobody had rotated.

None of those are Vanta failures. Vanta was doing exactly what it was designed to do. They are code-and-config failures that compliance automation was not built to find.

How we run them together

On our tenant and on every client we onboard, the split looks like this. Vanta or Drata owns evidence collection, policy management, control mapping, and the audit-ready export. AiT Audit owns the code-level deep audit: IAM blast-radius analysis on every Cloud Run service, webhook signature verification across every endpoint, OAuth scope minimality checks against actual code call patterns, AI-system controls mapped against OWASP LLM Top 10 and NIST AI 600-1, and continuous secret-scanning across Gist, gist-like surfaces, and historical commit ranges.

The output of AiT Audit feeds back into Vanta as evidence. Vanta gets richer. The auditor gets a more defensible report. The client gets coverage on the surface that the compliance platforms were not built for.

What to do this quarter

1. Audit the audit. Ask your compliance platform vendor (Vanta, Drata, Secureframe, anyone) for the explicit list of what their integrations cover and, more importantly, what they do not cover. Get it in writing.
2. Map your AI surface. List every model endpoint, every embeddings index, every fine-tuned model, every prompt-construction service. Cross-check against OWASP LLM Top 10. Most teams cannot fill in the spreadsheet.
3. Run a OAuth scope audit on every integration with customer data access. The 2025 SaaS supply-chain breaches will be 2026 SaaS supply-chain breaches and 2027 SaaS supply-chain breaches. The shape will not change.
4. Sign your SOC 2 with eyes open. A clean Type II is necessary. It is not the same thing as being secure.

The bottom line

The bottom line: Vanta and Drata won the evidence-collection war. They are not the right tool to find the next breach in your stack, and they were never designed to be. The MSPs telling clients otherwise are oversimplifying for the renewal call. The right answer in 2026 is both, with the boundary clearly drawn.