Two moments test a security record. One is routine: an auditor picks a sample and asks you to show the approval, the review, or the deprovisioning for each item. The other is not: an incident, after which someone reconstructs what happened, what was known, and what was done, from whatever exists.
Both are judged on what you can produce, not on what you did. A control that operated perfectly but left no trace is, for both purposes, a control that did not operate. Retention periods and legal holds are questions for counsel; this note covers what to keep and how to keep it trustworthy.
Four kinds of record
1. Decisions
Risk acceptances, exceptions, policy approvals, and the decisions that shaped the program: why a tool was retired, why a control was deferred. NIST CSF 2.0 makes recording exceptions an outcome (ID.RA-07). Keep the decision, the reason, who made it, and the date. The risk acceptance brief and the decision log on this desk cover the format.
2. Control operation
Evidence that controls ran: access review completions with the reviewer's name, joiner and leaver tickets with timestamps, vulnerability scan results and remediation tickets, backup and restore test results, training records. This is what auditors sample, and the identity audit piece describes what they open first.
3. Logs
System, authentication, and security tool logs. CSF 2.0 lists log records being generated and made available for continuous monitoring as an outcome (PR.PS-04). For a published reference on duration, OMB memorandum M-21-31 set US federal agencies' highest-criticality log categories at 12 months in active storage and 18 months in cold storage. That memorandum binds federal agencies only, but it is one of the few public retention figures written by someone who had to justify it.
4. Incident records
Timeline, decisions, actions taken, evidence collected, and communications. NIST SP 800-61 Rev. 3, published in April 2025, makes two points worth taking literally. Collected incident data is evidence even when formal chain-of-custody handling is not performed. And evidence should be collected and retained in line with the organization's evidence preservation procedures and data retention policies, considering the possibility of prosecution and the cost of keeping both the data and the means to read it later.
That last point is easy to miss. A forensic image is useless in three years if nobody can still open the format.
Making the record trustworthy
A record is only as useful as the confidence that it was not changed afterwards. CSF 2.0 states it directly for investigations: actions performed are recorded, and the records' integrity and provenance are preserved (RS.AN-06), and incident data and metadata are collected with integrity and provenance preserved (RS.AN-07).
In practice:
- Write once where possible. Logs and incident evidence go to storage that the people being investigated cannot alter or delete.
- Timestamp at creation. A decision recorded on the day it was made carries more weight than one reconstructed a month later. If you do write something up later, say when.
- Keep the original alongside the summary. The summary is for reading; the original is for checking.
- Record who had access. Especially for incident evidence, keep a short log of who handled what and when.
The incident timeline is a record too
During an incident, start a timeline document immediately and keep it running. Every entry gets a time, who, and what: detection, each decision, each containment action, each external contact.
For US public companies, one entry carries particular weight: when and how materiality was determined. The SEC requires disclosure of a material cybersecurity incident on Form 8-K generally within four business days after the registrant determines it is material, so the date of that determination, and the basis for it, will be examined later. How materiality is decided is a matter for counsel and the disclosure committee; the security leader's job is to make sure the facts they relied on, and when they had them, are in the timeline.
What not to keep
Retention has a cost, and keeping everything forever is its own exposure. Personal data in logs and evidence falls under privacy obligations, and counsel may set limits as well as minimums. Decide retention per record type, write it down, and apply it consistently. A retention schedule applied consistently is defensible; ad hoc deletion is not, especially once a dispute or investigation is foreseeable, when counsel may issue a hold.
A short retention map
| Record | Keep | Notes |
|---|---|---|
| Risk acceptances and exceptions | Life of the risk, plus your policy period | Include expired ones |
| Access reviews and leaver records | At least one full audit cycle | Auditors sample the period under review |
| Security and authentication logs | Set by policy and counsel | M-21-31 is one published reference point |
| Incident records and evidence | Set by counsel per incident | Keep the means to read the format |
| Board and committee papers | Per corporate records policy | Minutes record what the board was told |
Fill the "keep" column with your own policy and counsel's advice. The structure matters more than the numbers.
Next steps
- List the four record types and, for each, where it lives today and who can delete it.
- Move logs and incident evidence to write-once or access-restricted storage.
- Create an incident timeline template and make starting it the first step of the response plan.
- Agree retention per record type with counsel and write it down.
- For response mechanics, see the ransomware readiness guide, and for templates see /ciso/templates/.