ISO/IEC 27001 is a certification that an organisation runs an information security management system that meets the standard. SOC 2 is an auditor's attestation report on specific controls, often tested over a period. For an AI agent platform, both tell a buyer about security of the platform. Neither tells them whether the agents behave correctly when they act.
What does each require, in plain words?
ISO/IEC 27001 sets requirements for an information security management system. The organisation defines its scope, assesses information security risks, chooses controls to treat them (with a reference set of controls in the standard's annex), records which controls apply and why in a statement of applicability, and runs a cycle of monitoring, internal audit, management review and improvement. An accredited certification body audits it and issues a certificate, which is then kept up through periodic surveillance audits.
SOC 2 is an attestation report from an independent auditor on a service organisation's controls, measured against criteria for security, availability, processing integrity, confidentiality and privacy. Security is always included; the others are chosen. A Type 1 report covers whether controls were suitably designed at a point in time. A Type 2 report also tests whether they operated effectively over a period. The report describes the system, lists the controls, sets out the auditor's tests and records any exceptions found.
The core difference: ISO/IEC 27001 says a management system exists and works. SOC 2 says these controls were tested in this way, with these results.
Who are they for?
Any organisation can seek ISO/IEC 27001 certification, and it is widely recognised internationally. The certificate is usually shareable and names the scope.
SOC 2 is aimed at service organisations whose customers rely on their controls. It is most common where buyers are in the United States, though buyers elsewhere ask for it too. The report is usually shared under a confidentiality agreement, because it details the controls.
Agent platform vendors often hold both. Buyers in regulated industries often ask for both.
The two also differ in timing. A certificate is valid across its cycle, with surveillance audits in between, while a Type 2 report looks back over a fixed period that ends before the report is issued. Whichever you read, check what has changed on the platform since, especially in how agents connect to customer systems.
What does each mean for a platform whose agents act?
Agent platforms add things that a conventional software service does not have: the agent's brief and instructions, connections to the customer's systems with real permissions, a model provider somewhere in the chain, and a stream of actions taken on the customer's behalf. Read either assurance with those in mind.
Scope. Does the ISO/IEC 27001 scope or the SOC 2 system description include the agent runtime and the tool connections, or only the web application and hosting? A scope that stops at the application layer misses where agents act. Our answer to whether ISO/IEC 27001 covers AI agents looks at this closely.
Model providers. In a SOC 2 report, a model provider is usually a subservice organisation. If it is carved out, the report does not test its controls, and you need separate assurance. In an ISO/IEC 27001 system, it should appear as a supplier with managed risks.
Change management. A change to an agent's brief or sources can change its behaviour as much as a code release. Check whether change controls cover those, not only code.
Action logs. Both may test logging. Ask whether logs capture each agent action and the decision behind it, not only user logins.
What neither tests. Neither assesses whether an agent's answers are grounded, whether it refuses out-of-scope requests, or whether it escalates when it should. SOC 2's processing integrity criterion is about complete and accurate processing, not the judgement of an agent. That evidence comes from certifying the agent itself, which we cover in our guide to what it means to certify an agent.
How do you get what you need from each?
| Requirement | Control | Evidence it leaves |
|---|---|---|
| Security management is real | Ask for the ISO/IEC 27001 certificate and scope statement | Certificate, scope, statement of applicability on request |
| Controls operate over time | Ask for a SOC 2 Type 2 report, not only Type 1 | Report with tests and exceptions |
| Agent runtime is in scope | Check scope and system description name the agent runtime and connections | Scope review notes |
| Model provider is covered | Identify carve-outs and obtain the provider's own assurance | Subservice list and their reports |
| Your side is done | Implement complementary controls the report assigns to you | Mapping of your controls to the report's list |
| Agent behaviour is proven | Certify each agent capability on your own cases | Per-capability certification results |
What will your auditor or supervisor ask?
- Which assurance did you rely on when choosing this platform, and what did it cover?
- Did the scope include the parts of the platform where agents take actions?
- How did you handle carved-out subservice organisations?
- Which complementary controls are yours, and do you run them?
- What evidence shows the agents themselves perform as intended?
What do they not cover?
Neither standard governs AI risk as such. ISO/IEC 42001 sets requirements for an AI management system, compared in our pieces on ISO/IEC 42001 vs ISO/IEC 27001 and ISO/IEC 42001 vs SOC 2. ISO/IEC 23894 gives guidance on AI risk management; see how ISO/IEC 23894 applies to agents. Neither replaces sector rules such as DORA for ICT third-party risk in EU finance, or data protection law.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
Questions people ask
Is SOC 2 better than ISO/IEC 27001 for an agent platform?
Neither is better; they answer different questions. A Type 2 report shows tested controls and exceptions, while the certificate shows a management system that keeps improving. Many buyers want both.
Can an agent platform's SOC 2 report cover agent accuracy?
Processing integrity covers complete and accurate processing of data, which helps. It does not test an agent's judgement on your cases, so you still need certification of the agent's capabilities.
Should a buyer accept a Type 1 SOC 2 report?
As a starting point for a new vendor, perhaps. It shows design at one moment, not operation over time, so ask when the Type 2 will follow.