<- all tokdocs

SOC 2 and Enterprise Security Reviews: What Buyers Ask Vendors

By , founder of Mavric and Apso

View on TikTok ->

An enterprise security review checks whether a vendor can explain where customer data lives, who can reach it, what gets logged and what happens when something fails. A TikTok from @sd141414 says buyers start by asking someone on your team to explain the system. The four questions it lists are real control requirements, but the review starts with documents you write, and buyers do ask about your code in writing.

What buyers ask in a security review

The video lists four things a buyer wants explained: "Where the data is stored, who can access it, what gets logged, and what happens when a piece of it fails." Each one maps to a published criterion.

Data location. CAIQ v4.0.2 question DSP-19.1 asks: "Are processes, procedures, and technical measures defined and implemented to specify and document physical data locations, including locales where data is processed or backed up?" ISO/IEC 27001:2022 covers the same ground in Annex A control A.5.23, "Information security for use of cloud services."

Access. SOC 2 criterion CC6.1 in the AICPA 2017 Trust Services Criteria reads: "The entity implements logical access security software, infrastructure, and architectures over protected information assets to protect them from security events to meet the entity's objectives." ISO 27001:2022 carries the same requirement as A.8.3, "Information access restriction."

Logging. SOC 2 CC7.2 requires that "The entity monitors system components and the operation of those components for anomalies," with a point of focus on "logging of unusual system activities." CAIQ LOG-04.1 asks: "Is access to audit logs restricted to authorized personnel, and are records maintained to provide unique access accountability?" ISO 27001:2022 A.8.15 is titled "Logging."

Failure behavior. SOC 2 criterion A1.2, in the availability category, covers "environmental protections, software, data backup processes, and recovery infrastructure."

The video's follow-up question, "Can one customer's data ever reach another customer?", is a plain-language version of CAIQ IVS-06.1, which asks whether tenant access is "appropriately segmented, segregated, monitored, and restricted from other tenants." On what buyers ask, the video is accurate.

How does an enterprise security review start?

The video opens with "Enterprise buyers don't ask to see your code. They ask someone on your team to explain the system. That's what a security review is."

In practice the review starts with a document request. The first document is usually a SOC 2 Type II report. That report includes a system description written by the vendor, and its content is governed by AICPA DC section 200, the 2018 Description Criteria. DC section 200 has nine criteria. DC1 covers the services provided and how customers use the system. DC2 covers the principal service commitments and system requirements. Another criterion requires a statement of the system's boundaries, meaning what is in scope and what is not.

Alongside the report, or in place of it, the buyer sends a standard questionnaire:

  • CAIQ v4.1.0 has 261 yes-or-no questions mapped to 207 controls across 17 domains.
  • HECVAT 4, released by EDUCAUSE on 10 February 2025 and now at revision 4.1.6, has 321 questions across 7 sections.
  • The Shared Assessments SIG comes in a Lite form of roughly 126 questions and a Core form of roughly 855 questions across 21 risk domains.

All of that is written and finished before anyone schedules an architecture call. The live conversation in the video does happen, but as an escalation. It comes up when written answers contradict each other, when a control is marked as the customer's responsibility and the buyer expected the vendor to own it, or when an answer is a bare "Yes" with no description of how the control works.

Do enterprise buyers ask to see your code?

They will not read your repository. CAIQ, HECVAT and the SIG are control attestations, and SOC 2 is an auditor's opinion on controls rather than a code audit.

The questionnaires do ask about your code in writing:

  • CAIQ AIS-05.1 asks whether the testing strategy sets "criteria to accept new information systems, upgrades, and new versions."
  • AIS-05.2 asks: "Is testing automated when applicable and possible?"
  • AIS-06.1 asks whether you deploy "application code in a secure, standardized, and compliant manner."
  • AIS-07.1 asks: "Are application security vulnerabilities remediated following defined processes?"
  • TVM-06.1 asks about "the periodic performance of penetration testing by independent third parties."

Large buyers also ask for the code itself through source code escrow. The vendor deposits source with a third-party agent, and the agent releases it to the customer on defined triggers such as the vendor's bankruptcy or failure to maintain the product. ISACA's 2022 guidance describes escrow as something to raise in the RFP at procurement and then settle in the contract.

Are AI agents writing most of the code?

The video's turn is the line "But now that agents are writing most of it, creating code faster than anyone can read, a team can ship a working product that no one can explain."

Anthropic's 2026 Agentic Coding Trends Report gives a smaller figure. It says developers "use AI in roughly 60% of their work" but "report being able to 'fully delegate' only 0-20% of tasks." The report describes the shift as one "from writing code to reviewing, directing, and validating AI-generated code," and says "the human role remains central." It does predict that organizations will use agents to review large volumes of AI-generated code, but that is a forecast in the document rather than a measurement.

The video's broader point still holds. As more implementation is delegated, fewer people hold the full picture of the system, and the security review is where that shows. The "most of it" figure is the part that does not match the source.

Does hesitating on a question count against you?

The video says "If your answer is, I'd have to check, the buyer hears yes." No framework scores verbal hesitation. SOC 2 scores whether a control operated over an observation period. CAIQ and HECVAT score written answers. No criterion in the Trust Services Criteria, CCM v4 or HECVAT 4 turns "I would have to check" into a finding. An unanswered architecture question usually produces a follow-up request or a remediation item.

Bottom line

The four questions in the video are the right ones, and each maps to a SOC 2, CAIQ or ISO 27001 control. The review starts with a vendor-written SOC 2 system description and a questionnaire, and the architecture call comes later if gaps show up. Buyers ask about testing, deployment and patching in writing, and large deals can require source code escrow.

Sources

Published October 5, 2026. Based on a TikTok by @sd141414. Full video notes (markdown)