What you will be able to do
- Explain where Snowflake's certifications end and the customer's own compliance work begins
- Choose the edition and agreement a PHI or cardholder-data workload needs before any data is loaded
- Match masking, row access and access-history controls to specific HIPAA, PCI DSS, GDPR and CCPA requirements
Key concept
Shared compliance responsibility — Snowflake's certifications and attestations cover the service it runs. Deciding which regulations apply to your data, configuring the account to meet them, and producing evidence that you did all stay with you.
1.What a Snowflake certification does and does not cover
A Snowflake certification shows that the platform is fit to host regulated data. It does not show that your account is configured to protect it. Snowflake holds a long list of attestations, including SOC 1 and SOC 2 Type II, ISO-27001, ISO-27017 and ISO-27018, HITRUST CSF, PCI DSS, FedRAMP (Moderate and High), ITAR and IRAP (Protected). Each one is evidence about Snowflake's own controls. An auditor reviewing your workload still asks about your users, your roles, your policies and your logs.
PCI DSS shows the split clearly. Snowflake is a Level 1 Service Provider under PCI DSS version 3.2.1 and is assessed every year by a Qualified Security Assessor. That lets customers store, process or transmit cardholder data in Snowflake. Some PCI requirements remain the customer's job, though, and Snowflake publishes a PCI Shared Responsibility Matrix, available on request, that shows which ones. Start any PCI scoping exercise from that matrix. A requirement on your side of the line is not met just because the service is certified.
The built-in tools follow the same rule. Snowflake's sensitive-data classification groups detected columns into compliance categories such as GDPR. Snowflake describes these categories as an out-of-the-box toolkit that is not exhaustive. They help you find regulated data. They do not decide for you which laws apply to it.
Checkpoint 1 of 9· Check yourself
A retailer plans to load cardholder data into Snowflake. A project lead says: "Snowflake is a PCI DSS Level 1 Service Provider assessed every year by a QSA, so our PCI obligations are covered." What is the right response?
Snowflake's certification is evidence about Snowflake's own controls. Requirements on the customer's side of the line are not met just because the service is certified, and the Shared Responsibility Matrix shows which requirements those are.
“there are PCI compliance responsibilities that fall to customers outside of those managed by Snowflake”Source: docs.snowflake.com
2.HIPAA and PCI DSS workloads start with the edition
For PHI, two things must both be true before the first row is loaded. The account must be on Business Critical edition or higher, and a signed business associate agreement (BAA) must exist between your organisation and Snowflake Inc. HIPAA and HITRUST CSF require the BAA before any PHI is stored, so the order matters. Signing it after a pilot load means PHI was already held without it.
Options that sound strong on security do not change edition eligibility. Periodic rekeying is available from Enterprise edition, but PHI support is not. Masking policies protect individual columns, but they do not bring a Standard account into HIPAA scope. Tri-Secret Secure cannot be the fix for a Standard account either, because it is a Business Critical feature. PCI DSS support and FedRAMP or ITAR public-sector workloads sit at the same Business Critical tier.
| Feature | Standard | Enterprise | Business Critical |
|---|---|---|---|
| Automatic encryption of all data | Yes | Yes | Yes |
| Support for multi-factor authentication | Yes | Yes | Yes |
| Periodic rekeying of encrypted data | No | Yes | Yes |
| Column-level Security (masking policies) | No | Yes | Yes |
| Row-level Security (row access policies) | No | Yes | Yes |
| Audit user access history through ACCESS_HISTORY | No | Yes | Yes |
| Customer-managed encryption keys through Tri-Secret Secure | No | No | Yes |
| Support for PHI data (HIPAA and HITRUST CSF) | No | No | Yes |
| Support for PCI DSS | No | No | Yes |
The edition also limits what you can show later. A Standard account has no masking policies, no row access policies and no ACCESS_HISTORY view, so it cannot produce the access-audit evidence that GDPR and CCPA reviews rely on. The account's edition is itself evidence, and you can read it with SQL.
SELECT edition FROM SNOWFLAKE.ORGANIZATION_USAGE.ACCOUNTS WHERE account_name = CURRENT_ACCOUNT();Checkpoint 2 of 9· Check yourself
For a GDPR review, an auditor asks for evidence of which user wrote to a table holding personal data, and when. The account runs on Standard edition. What is the right plan?
Access-audit evidence of who wrote to a table and when comes from ACCESS_HISTORY, which starts at Enterprise edition. Business Critical is not required for it, and Standard has neither ACCESS_HISTORY nor masking policies, so the edition limits what evidence the account can produce.
“Identify the Snowflake user who performed a write operation on a table or stage and when the write operation occurred”Source: docs.snowflake.com
Checkpoint 3 of 9· Exam question
A hospital analytics team must load protected health information (PHI) into a new Snowflake account under a HIPAA business associate agreement. The account was created as Standard edition. What must the security engineer do before PHI is loaded?
Correct answer: A — Move the account to Business Critical edition and sign a business associate agreement with Snowflake before PHI is loaded
- A. Correct. Snowflake supports PHI under HIPAA and HITRUST only on Business Critical (and higher) editions, and a BAA must be in place before regulated data is loaded.
- B. Incorrect. Tri-Secret Secure is not available on Standard edition, and a customer-managed key does not by itself make an account eligible for PHI.
- C. Incorrect. Masking policies are a column-level control and cannot change the edition-level support boundary; masking does not place a Standard account in HIPAA scope.
- D. Incorrect. Periodic rekeying is an Enterprise feature, but the minimum edition for PHI support is Business Critical, not Enterprise.
3.Encryption and MFA as the baseline every regulation expects
Every regulation in scope expects data to be encrypted at rest and in transit, and Snowflake does this on every edition without any setup. All customer data at rest is encrypted, and traffic to and from the service uses TLS. Files uploaded to an internal stage are encrypted on the client before they leave the user's machine. For external stages Snowflake recommends client-side encryption, and it encrypts the data as soon as it is loaded into a table. Because none of this is configurable, your auditor's evidence is the platform documentation rather than an account setting.
Customer control of keys is a separate requirement, and it is the one that needs a decision from you. Where a regulator or internal policy requires the customer to hold a key Snowflake cannot use alone, the answer is Tri-Secret Secure customer-managed key. It combines your key with Snowflake's and requires Business Critical edition. Periodic rekeying from Enterprise edition up limits how long any one key protects data. It does not give you control of the key.
MFA is the other baseline control, and auditors want proof that it is enforced, not just available. Enrollment mechanics are covered in MFA enrollment, authenticators and passkeys. For compliance, the evidence is an authentication policy that requires every password-using human user to enroll. The Trust Center's always-on Security Essentials scanners check for exactly that policy and for actual enrollment, so a clean result from them is evidence an assessor can accept.
Checkpoint 4 of 9· Check yourself
An internal auditor asks the data team to prove that data at rest and in transit is encrypted in their Enterprise edition Snowflake account. What should the team provide?
Encryption at rest and TLS in transit are automatic on every edition and cannot be configured, so the evidence is the platform documentation. Rekeying limits how long any one key protects data, and Tri-Secret Secure gives the customer control of a key. Neither is what turns encryption on.
“In Snowflake, all customer data at rest is encrypted and encrypted with TLS in transit to/from the Snowflake service.”Source: docs.snowflake.com
4.Masking personal data, PHI and cardholder data with tags
Data minimisation under GDPR, the minimum-necessary handling of PHI, and restricting display of cardholder data all come down to one requirement: most roles should see a masked value, and a named few should see clear text. Dynamic Data Masking policies (column-level security) do this at query time. What matters for compliance is who decides. A security or privacy officer chooses which columns to protect, not the object owner. That separation of duties is what an auditor wants to see.
At scale, set the masking policy on a tag instead of on each column. Suppose a classification job tags email columns across 40 databases and new tables arrive every week. If you set one masking policy on that tag, every column carrying the tag is protected as soon as its data type matches the policy signature. Set the same tag on a schema or database, and columns in newly added tables are covered automatically. The policy body holds the role check, for example returning clear text only when the role is the compliance officer's role.
ALTER TAG my_tag SET MASKING POLICY my_masking_policy;The other approaches all need ongoing work. A nightly procedure that runs ALTER TABLE ... SET MASKING POLICY column by column leaves a gap between a table's creation and the next run, and its failures need someone to review them. Wrapping every table in a secure view with a CASE on CURRENT_ROLE() means writing and maintaining one view per table. A row access policy on the tag is the wrong control entirely: it removes whole rows instead of hiding the email value.
Two rules affect the design. A tag can hold one masking policy per data type, so cover STRING, NUMBER and timestamp columns with one generic policy each. And if a column has both a directly assigned masking policy and a tag-based one, the directly assigned policy wins. A local exception therefore overrides the central rule. For evidence, POLICY_REFERENCES lists the policy name, tag name and policy status for each protected column.
SELECT *
FROM TABLE(INFORMATION_SCHEMA.POLICY_REFERENCES(
REF_ENTITY_DOMAIN => 'TABLE',
REF_ENTITY_NAME => 'my_database.my_schema.my_table'
));Checkpoint 5 of 9· Check yourself
In a HIPAA readiness review, the assessor asks who decides which PHI columns get masked. Which arrangement shows the separation of duties the assessor wants to see?
Dynamic Data Masking separates duties because a security or privacy officer, not the object owner, decides which columns are protected. Letting owners decide, through their own policies or per-table secure views, removes that separation. A privileged admin role is not the separation the auditor is looking for either.
“A security or privacy officer decides which columns to protect, not the object owner.”Source: docs.snowflake.com
Checkpoint 6 of 9· Exam question
A health insurer tags every column that holds personal email addresses with the object tag PII_EMAIL across 40 databases, and new tables appear weekly. To support GDPR data minimisation, only the COMPLIANCE_OFFICER role may see clear text. Which design needs the least ongoing maintenance?
Correct answer: A — Create a masking policy that tests IS_ROLE_IN_SESSION('COMPLIANCE_OFFICER') and set it on the PII_EMAIL tag so tagged columns inherit it
- A. Correct. A tag-based masking policy is applied automatically to every column that carries the tag, including columns tagged on tables created later, so no per-column work is needed.
- B. Incorrect. A scheduled procedure can eventually achieve the same result, but it leaves a gap between tagging and protection and needs constant upkeep and failure handling.
- C. Incorrect. Secure views must be built and kept in sync for every table, and the underlying tables stay exposed to anyone granted access to them.
- D. Incorrect. Row access policies cannot be assigned to tags and filter rows rather than masking column values, so this does not satisfy the requirement.
5.Regional processing restrictions need row access, not masking
Rules about where data may be processed, such as keeping EU residents' orders visible only to the EU team, limit which records a person may see, not how a value is displayed. A row access policy is the control for this. It decides which rows a query returns, based on the values in one or more columns. Picture one orders table with a REGION column. The policy compares REGION against the invoking role and passes only matching rows. If the role-to-region assignments live in a lookup table the policy reads, a monthly change of team membership is a data update, not a policy rewrite.
| Control | What it actually restricts | Fit for regional row restriction |
|---|---|---|
| Row-level Security (row access policies) | Which rows are returned in a query result | Yes: non-matching rows are filtered out |
| Column-level Security (masking policies) | How a column's value is displayed | No: masking REGION still returns every order |
| Network policies | Site access by user IP address | No: limits where users sign in from, not which rows they read |
| Aggregation policies | Queries must aggregate data to return results | No: blocks row-level output for everyone and does not split by region |
Row access policies can also be set on a tag, using the same ALTER TAG pattern as masking. Any table carrying the tag, directly or through its schema or database, is then filtered. One limit is worth planning around: the same column cannot appear in both a masking policy signature and a row access policy signature at once. So a REGION column that drives row filtering should not also be masked.
Checkpoint 7 of 9· Check yourself
Under a data-residency rule, analysts on the EU team may see only EU customers' orders in a shared ORDERS table that has a REGION column, and team membership changes every month. Which design meets the rule with the least ongoing work?
The rule limits which records a person may see, and only a row access policy controls which rows a query returns. A masking policy only changes how a value is displayed, so every order is still returned. A network policy limits where users sign in from, and an aggregation policy blocks row-level output for everyone.
“Dynamic Data Masking is a Column-level Security feature that uses masking policies to selectively mask plain-text data in table and view columns at query time.”Source: docs.snowflake.com
Checkpoint 8 of 9· Exam question
A retailer keeps customer orders for EU and US residents in one table with a REGION column. To honour regional processing restrictions, the EU analyst team must see only EU rows and the US team only US rows, and team assignments change monthly. Which approach fits best?
Correct answer: A — Attach a row access policy that looks up the invoking role in a role-to-region mapping table and passes only matching REGION rows
- A. Correct. A row access policy driven by a mapping table filters rows per role at query time, and monthly reassignments only require updating the mapping table.
- B. Incorrect. A masking policy changes the value of one column, so the rows themselves would still be returned to the wrong team.
- C. Incorrect. Network policies decide where a connection may originate and have no effect on which rows a role can read.
- D. Incorrect. An aggregation policy forces aggregate queries over minimum group sizes but does not partition data between teams by region.
6.Answering GDPR and CCPA questions about who touched a person's data
GDPR and CCPA reviews ask questions that need records at the level of each statement: who read this person's data, who changed it, and where copies of it went. The ACCESS_HISTORY view in ACCOUNT_USAGE answers them. Each row links a user, a query, the objects and columns read or written, and any policies enforced. Snowflake states that these records facilitate regulatory compliance auditing. How direct and base objects are recorded is covered in ACCESS_HISTORY (direct vs base objects, columns, writes).
Three features of the view map onto specific compliance questions:
- Write auditing identifies the user who performed a write on a table or stage and when it happened, which is the trail GDPR and CCPA reviews ask for.
- Column lineage in objects_modified follows a sensitive column through INSERT, MERGE and CTAS into derived tables. That lets a privacy officer count how many objects hold copies of a column, which is what a GDPR erasure or minimisation review needs.
- The policies_referenced column records which masking or row access policies were enforced on each query, so you can show that protected data was read under its policy without joining POLICY_REFERENCES to QUERY_HISTORY.
Account-level access history costs nothing extra. The organisation-wide ORGANIZATION_USAGE.ACCESS_HISTORY view, which exists only in the organization account, is a premium view billed for serverless compute and storage. Not every statement in QUERY_HISTORY appears in ACCESS_HISTORY, because the structure of the SQL decides whether a record is written. Test the specific statement types you rely on before you promise an auditor complete coverage.
Checkpoint 9 of 9· Match them up
A privacy officer is preparing for a GDPR and CCPA review. Match each part of Snowflake's access history to the question it answers.
Tap a term, then the definition that fits it.
Each compliance question maps to a specific feature. Write auditing records who wrote and when. Column lineage counts the derived copies of a sensitive column. policies_referenced shows which policies were enforced on each query. The organisation-wide view combines all accounts at extra serverless cost, while the account-level view costs nothing extra.
“Data privacy officers can quickly determine the object count (e.g. 1 table, 2 views) of a column containing sensitive data.”Source: docs.snowflake.com
Sources11
Sources
(11)
Every claim above is drawn from one of these pages, quoted as it was written on the date shown.
- 1.
“there are PCI compliance responsibilities that fall to customers outside of those managed by Snowflake”
↩︎ What a Snowflake certification does and does not cover“Snowflake is a Level 1 Service Provider compliant under PCI DSS version 3.2.1”
↩︎ What a Snowflake certification does and does not cover“customers can request the Snowflake PCI Shared Responsibility Matrix”
↩︎ What a Snowflake certification does and does not cover - 2.
“FedRAMP (Moderate and High)”
↩︎ What a Snowflake certification does and does not cover - 3.
“You are solely responsible for determining which regulations or laws apply to your data and ensuring your compliance with the applicable regulations or laws”
↩︎ What a Snowflake certification does and does not cover“You are solely responsible for determining which regulations or laws apply to your data and ensuring your compliance with the applicable regulations or laws”
↩︎ Key concept - 4.
“before any PHI data can be stored in Snowflake, a signed business associate agreement (BAA) must be in place”
↩︎ HIPAA and PCI DSS workloads start with the edition“Periodic rekeying of encrypted data for increased protection.”
↩︎ HIPAA and PCI DSS workloads start with the edition“Customer-managed encryption keys through Tri-Secret Secure.”
↩︎ Encryption and MFA as the baseline every regulation expects“before any PHI data can be stored in Snowflake, a signed business associate agreement (BAA) must be in place”
↩︎ Exam trap 1 - 5.
“Support for PHI data (in compliance with HIPAA and HITRUST CSF regulations) — requires Business Critical Edition (or higher).”
↩︎ HIPAA and PCI DSS workloads start with the edition - 6.
“In Snowflake, all customer data at rest is encrypted and encrypted with TLS in transit to/from the Snowflake service.”
↩︎ Encryption and MFA as the baseline every regulation expects - 7.
“You have an authentication policy that enforces all human users to enroll in MFA if they use passwords to authenticate.”
↩︎ Encryption and MFA as the baseline every regulation expects - 8.
“A security or privacy officer decides which columns to protect, not the object owner.”
↩︎ Masking personal data, PHI and cardholder data with tags“Dynamic Data Masking is a Column-level Security feature that uses masking policies to selectively mask plain-text data in table and view columns at query time.”
↩︎ Checkpoint - 9.
“the columns in all newly added tables and views are automatically protected when the column data type matches the masking policy data type”
↩︎ Masking personal data, PHI and cardholder data with tags“A tag can support one masking policy for each data type.”
↩︎ Masking personal data, PHI and cardholder data with tags“the same column cannot be specified in both a masking policy signature and a row access policy signature at the same time”
↩︎ Regional processing restrictions need row access, not masking“the masking policy that is directly assigned to the column takes precedence over the tag-based masking policy”
↩︎ Exam trap 2 - 10.
“A row access policy is used to control which rows are returned in a query result.”
↩︎ Regional processing restrictions need row access, not masking“A row access policy is used to control which rows are returned in a query result.”
↩︎ Exam trap 3 - 11.
“when the write operation occurred to meet compliance regulations, such as GDPR and CCPA”
↩︎ Answering GDPR and CCPA questions about who touched a person's data“data privacy officers can prove how they satisfy regulatory compliance standards”
↩︎ Answering GDPR and CCPA questions about who touched a person's data“There is no additional cost associated with the ACCOUNT_USAGE.ACCESS_HISTORY view.”
↩︎ Answering GDPR and CCPA questions about who touched a person's data“The records in these views facilitate regulatory compliance auditing”
↩︎ Answering GDPR and CCPA questions about who touched a person's data“Identify the Snowflake user who performed a write operation on a table or stage and when the write operation occurred”
↩︎ Checkpoint“Data privacy officers can quickly determine the object count (e.g. 1 table, 2 views) of a column containing sensitive data.”
↩︎ Checkpoint