I've worked in pharma quality IT for more than 25 years, and I contributed to GAMP 5 Second Edition, which carries the same risk-based thinking. The question I hear most from validation leads is a fair one: "If I adopt CSA, what am I allowed to stop doing?" This page answers that, and it's just as clear about what you must keep doing. If you're new to the subject, start with my guide to computer system validation in pharma.
Where CSA comes from
FDA's Center for Devices and Radiological Health (CDRH) and Center for Biologics Evaluation and Research (CBER) published "Computer Software Assurance for Production and Quality System Software" as a draft in September 2022. It was finalised September 2025, updated February 2026 to align with the QMSR. The current version is titled "Computer Software Assurance for Production and Quality Management System Software", and it supersedes Section 6 of FDA's 2002 General Principles of Software Validation.
Scope matters here. The guidance is written for medical device manufacturers under 21 CFR Part 820. It doesn't mention 21 CFR Part 211, and drug manufacturers aren't its stated audience. Pharma companies apply the same principles through ISPE's GAMP 5 Second Edition (July 2022), which puts critical thinking at the centre of computer system compliance. So when a pharma site says "we've moved to CSA", it usually means it has adopted GAMP 5 Second Edition thinking and uses the FDA guidance as a reference for how testing and records can scale with risk.
FDA's CSA guidance carries no percentages. You'll see claims online that CSA cuts validation effort by a fixed amount, and none of them come from the guidance. The savings you get depend on how much low-value testing and paperwork you were doing before.
CSV vs CSA side by side
| Area | Traditional CSV as most sites practised it | CSA as described by FDA and GAMP 5 Second Edition |
|---|---|---|
| Purpose | Prove the system works, usually by producing a full document set | Gain confidence the system works for its intended use, with effort in proportion to risk |
| Regulatory basis | 21 CFR Part 11, 21 CFR 211.68, EU GMP Annex 11, GAMP 5 (2008) | The same regulations, plus FDA's CSA guidance (finalised September 2025, updated February 2026) and GAMP 5 Second Edition |
| Risk approach | Risk assessment often done, then every requirement tested to the same depth | Risk decides the depth: functions are split into high process risk and not high process risk |
| Testing types | Scripted test cases with expected results and step-by-step execution | Scripted testing for high-risk functions; unscripted testing (scenario, exploratory, error guessing) where risk is lower |
| Documentation | Large protocols, screenshots on every step, signatures on every page | Records sized to risk; a short note of what was tested, by whom, when, and the result can be enough for lower-risk functions |
| Evidence | Paper or PDF printouts and screenshots | Digital records such as system logs and audit trails, used directly as evidence |
| Effort focus | Spread evenly, often heaviest on easy-to-test, low-risk features | Concentrated on functions that affect patient safety, product quality and data integrity |
| Supplier use | Supplier documents reviewed, then much of the testing repeated in-house | Supplier testing assessed and used as a starting point, with in-house testing on what the supplier can't cover |
| Tools | Tools validated as if they were GxP systems themselves | Tools assured in line with their own risk, often with lighter evidence |
| Inspection posture | "Show me the binder" | "Show me your risk reasoning, and the evidence for the high-risk functions" |
The left column describes how CSV was commonly done, which was heavier than any regulation required. Most of the waste CSA removes was self-imposed.
What does not change
This is the part that gets lost in conference talks. Moving to CSA leaves every regulatory requirement in place.
21 CFR Part 11 still applies to electronic records and electronic signatures. You still need controls on access, audit trails, signature manifestation and record retention, and an inspector will still test them.
21 CFR 211.68 still requires that automatic, mechanical and electronic equipment used in drug manufacturing is routinely checked under a written programme, that only authorised people change master production and control records, and that inputs and outputs are checked for accuracy.
EU GMP Annex 11 still applies to computer systems used in GMP activities in Europe and in PIC/S member countries. The 2011 version remains in force. A revised draft went to public consultation from 7 July to 7 October 2025, and on 12 November 2026 the final text still hasn't been published, so treat the draft as a signal of direction and keep working to the 2011 text.
Data integrity expectations (ALCOA+, FDA's 2018 data integrity Q&A guidance, PIC/S PI 041-1) are unchanged. CSA lets you use the system's own audit trail as evidence, which only works if the audit trail is reliable in the first place.
Intended use is still the anchor. You must define what the system is for, in your process, before you can decide what's risky. If the intended use is vague, your risk assessment will be vague, and the whole CSA argument falls over.
3 common mistakes when moving to CSA
1. Treating CSA as permission to test less everywhere
I've seen teams read "unscripted testing" as "no testing". CSA moves effort. It takes effort away from low-risk functions and puts more of it on high-risk ones. If your total test effort on a batch release workflow went down, you've probably got the risk calls wrong.
2. Changing the templates and leaving the SOPs alone
A new lighter test template is easy to write. If your validation SOP still says every requirement needs a scripted test with a screenshot, your team will follow the SOP, and an auditor will ask why the records don't match it. Fix the procedure first, then the templates. My step-by-step guide to moving from CSV to CSA starts there for that reason.
3. Skipping the supplier assessment and still trusting the supplier
CSA lets you use supplier testing, and that's one of the biggest effort savings in a SaaS world. It depends on a documented supplier assessment covering their quality system, development practice and data integrity controls. Without that assessment, you've got no basis for relying on their evidence.
A worked example with a cloud eQMS deviation module
Here's how I'd approach one system. Take a SaaS eQMS where the site is rolling out the deviation and CAPA module. It's GAMP Category 4, configured, with no custom code. For cloud systems in general, my GxP cloud validation framework sets out who owns what.
First I'd write the intended use in 2 or 3 sentences: the module records deviations, routes them for QA impact assessment, links CAPAs, and captures the QA approval signature that closes the record. That signature is a Part 11 electronic signature, and the closure decision feeds batch disposition.
Then I'd split the functions by risk. QA approval and e-signature, the audit trail on deviation records, the impact assessment fields that drive batch decisions, and role-based access to approve are all high process risk. Report layouts, dashboard colours, email notification wording and search filters aren't.
For the high-risk functions, I'd run scripted tests with defined expected results, including negative tests: can a user without the QA role approve, and does the audit trail capture a changed impact rating with old value, new value, user and time? For the lower-risk functions, I'd run a recorded exploratory session by a trained user, with a short note of what was covered and anything found.
For the platform itself, I'd assess the supplier's SOC 2 report, release testing and change notification process, and rely on that evidence for core platform behaviour. I wouldn't repeat their regression testing.
The evidence would be the system's own audit trail exports and configuration records, with a summary report that explains each risk decision. The output is a smaller package where every page earns its place, and the reasoning is written down for the inspector.
How I help teams make the move
My method for this shift is called DRIVE, and I use it with validation and QA teams that want to move to CSA without creating inspection risk. If you want to see how it would apply to your own procedures, my CSV to CSA consulting page sets out how an engagement runs. Teams that are also bringing AI into GxP processes will find the same risk logic carried further in my guide to AI validation in GxP and in my book on validating AI in GxP.
Frequently asked questions
Is CSA replacing CSV?
CSA is a risk-based way of carrying out validation. The obligation to validate computer systems for intended use stays in 21 CFR 211.68, Part 11 and Annex 11. What changes is how much testing and paperwork you put behind each function.
Does the FDA CSA guidance apply to pharma?
It was written by CDRH and CBER for medical device production and quality management system software under 21 CFR Part 820. Pharma companies apply the same principles through GAMP 5 Second Edition. Check with your own regulatory function on how you reference it in your procedures.
When was the FDA CSA guidance finalised?
The draft came out in September 2022. It was finalised September 2025, updated February 2026 to align with the QMSR. The February 2026 version replaces the September 2025 final text.
Do I still need a validation plan and summary report under CSA?
Yes, in my view, for any GxP system that matters. They get shorter and more focused. The plan explains intended use and the risk approach, and the summary report records what you did and why it's enough.
Can I use unscripted testing for Part 11 functions?
I wouldn't, for electronic signatures, audit trails or access control on records that support release decisions. Those are high-risk functions and they deserve scripted tests with clear expected results. Unscripted testing fits the lower-risk parts of the same system.
Will EU inspectors accept CSA?
Annex 11 has asked for risk-based validation since 2011, so the thinking isn't new to European inspectors. In my experience what they look for is a documented rationale for each risk decision. Use GAMP 5 Second Edition language in your procedures and you'll be on familiar ground.
Train your team on CSA with your own systems
I run in-house CSA training built around your SOPs and a live system of yours, so the team leaves able to classify functions and defend the calls. If you'd rather talk it through first, book a 30-minute call.
Sources
Last checked: 12 November 2026. I checked the sources listed below on that date.
- FDA, Computer Software Assurance for Production and Quality Management System Software (guidance page, February 2026)
- FDA, CSA guidance text (PDF, February 2026)
- Federal Register, CSA draft guidance notice, 13 September 2022
- eCFR, 21 CFR Part 11
- eCFR, 21 CFR 211.68
- European Commission, EudraLex Volume 4 (Annex 11 listed as the January 2011 revision)
- European Commission, consultation on Chapter 4, Annex 11 and new Annex 22 (closed 7 October 2025)
- ISPE, GAMP 5 Second Edition
- FDA, Data Integrity and Compliance With Drug CGMP: Questions and Answers (December 2018)
- PIC/S, PI 041-1 Good Practices for Data Management and Integrity (1 July 2021)