SEC Cyber Disclosure AI Compliance Checklist

A disclosure committee opens a red incident ticket at 16:40 on Friday. The ticket names the provider and affected application. It also contains a screenshot of a blocked prompt. It lacks the discovery time, related events, data scope, business owner, and materiality decision authority. This SEC cyber disclosure AI compliance checklist grades that case remediate. Each check needs a named owner and dated test. It also needs source evidence and an open action, followed by the retest result.

TL;DR

• Apply the SEC rules to cybersecurity incidents and public-company disclosure processes. The rules contain no AI-specific control checklist.
• Grade eight checks as pass or out of scope, with remediate used when needed. Record scope, owner, evidence, and test date.
• Preserve the discovery and materiality timestamps separately. The four-business-day filing period generally follows the materiality determination.
• Credit AI gateway evidence only for authenticated HTTP traffic that crossed the gateway and only as incident or materiality support.

Check 1: confirm reporting scope and accountable owners

Owner: securities counsel and the disclosure committee, supported by cybersecurity and finance, plus investor relations and internal audit.

Pass criteria: the company documents which entities and Exchange Act reports are in scope. Domestic registrants map material incidents to Form 8-K Item 1.05 and annual cyber disclosures to Regulation S-K Item 106. Foreign private issuers map comparable requirements to Forms 6-K and 20-F. The procedure identifies who determines materiality and approves disclosure, plus who requests a delay and files amendments.

Evidence: reporting-entity inventory and responsibility matrix. Add the disclosure controls and filing calendar, along with one completed test.

The SEC's 2023 adopting release provides the authoritative scope and rule text for public-company cybersecurity disclosure. It creates no separate AI compliance regime.

Remediate when: an AI compliance owner is expected to make a securities-law decision without named legal and disclosure authority, or a subsidiary incident has no route into the registrant's process.

Check 2: define and reconcile the cyber incident population

Owner: the incident-response leader. Legal and privacy teams support classification and reconciliation, along with business continuity and application teams.

Pass criteria: intake uses the rule's cybersecurity incident definition and covers unauthorized occurrences, including a series of related unauthorized occurrences, on or through information resources that jeopardize availability or the confidentiality and integrity of those resources. The register includes investigated events later closed as benign and third-party notices. It also includes data exposure, system disruption, provider events, and recurring low-severity activity that may aggregate.

AI enters this check through actual event facts. Include unauthorized LLM calls and AI API abuse. Also include exposed prompt or response data and manipulated retrieval content, along with material service disruption when they meet the company's incident criteria. A model-quality defect with no cyber dimension belongs in its own risk process.

Evidence: incident standard and frozen register export. Retain the source queries and reconciliation results, along with discrepancy tickets and aggregation decisions. Compare the register with SOC cases and legal and privacy matters, plus continuity cases, API-management events, and provider notices.

Remediate when: the population contains only confirmed breaches, or related AI events stay split across application teams with no aggregation review.

Check 3: escalate facts for materiality without unreasonable delay

Owner: the disclosure committee or other authorized materiality body. Legal and finance provide the factual record, along with security, operations, and business owners.

Pass criteria: discovery triggers a documented escalation path. The team gathers scope and timing, plus affected systems and operational consequences. The record also covers financial consequences, customer and vendor effects, reputation, litigation, and regulatory exposure. Decision makers receive updates as estimates change. The record shows when the company discovered the incident and determined materiality.

The SEC's Division of Corporation Finance said in its May 21, 2024 staff statement that materiality analysis should consider qualitative factors alongside quantitative factors. It names customer or vendor relationships and reputation as examples. It also names competitiveness and litigation, along with regulatory action. The staff statement has no legal force and creates no new duty.

Evidence: escalation timeline and fact packets. Retain meeting records and the financial and legal analysis, along with the decision memo and later reassessments.

Remediate when: the procedure waits for a final forensic report before beginning materiality analysis, or the memo records one revenue estimate and omits qualitative effects.

Check 4: control Item 1.05 content and the filing clock

Owner: securities counsel with the disclosure committee. Finance and cybersecurity support drafting and approval, along with investor relations and the corporate secretary.

Pass criteria: after the registrant determines a cybersecurity incident is material, the process calculates the general four-business-day deadline. The draft describes material aspects of nature and scope. It also describes timing and material impact or reasonably likely material impact. Unknown information is tracked for amendment. Technical detail is limited when its disclosure would impede response or remediation.

Evidence: dated materiality approval and deadline calculation. Retain the controlled drafts and source support, along with final authorization and EDGAR acceptance. Track amendments separately. The SEC fact sheet summarizes the filing trigger and content. It also covers timing and annual requirements.

Remediate when: the incident ticket's creation date is treated as the materiality date, the filing deadline lives in one person's inbox, or the disclosure draft cannot trace each material statement to a current source.

I would fail this check if counsel receives a polished narrative but no preserved chronology. Four business days leaves little room to discover that three teams used different incident clocks.

Check 5: separate required, voluntary, and delayed filings

Owner: securities counsel, supported by the disclosure committee. The company's law-enforcement liaison also supports this work.

Pass criteria: the procedure distinguishes a required Item 1.05 filing for an incident determined material from a voluntary Form 8-K disclosure under another item, such as Item 8.01, for an incident pending materiality review or determined immaterial. A later materiality determination starts the Item 1.05 filing workflow. Delay occurs only after the United States Attorney General makes the required national-security or public-safety determination and notifies the SEC in writing.

The SEC's Form 8-K compliance and disclosure interpretations state that requesting a delay leaves the original filing obligation unchanged unless the Attorney General's written determination arrives before the Form 8-K is due. They also address deadlines when a delay expires or ends early.

Evidence: filing classification memo and Item 8.01 and Item 1.05 register. Retain the later determination trigger and Attorney General determination, along with the delay calendar and counsel approval.

Remediate when: a pending request is treated as an approved delay, or voluntary disclosure under Item 8.01 closes the materiality workstream.

Check 6: support annual Item 106 disclosures with tests

Owner: the disclosure committee, supported by the CISO and enterprise risk. Management and the corporate secretary also support the work, along with internal audit.

Pass criteria: annual reporting describes the company's processes, if any, for assessing and identifying material risks from cybersecurity threats and managing them. It addresses material effects or reasonably likely material effects and board oversight. It also addresses management's role and expertise. Every statement in the draft links to operating evidence for the reporting period.

Evidence includes the cyber-risk process and enterprise-risk record. Add committee charters and board minutes. Also retain management reports, prior incidents, third-party monitoring, and open issues in a disclosure-support matrix. Select one statement and trace it. If the draft says management receives regular cyber reports, identify the dates and recipients. Retain the source metrics and decisions, along with overdue actions.

Remediate when: last year's Item 106 text rolls forward while reporting lines and provider dependencies changed. The same applies when board oversight or incident processes changed. A governance paragraph cannot pass through copy editing alone.

Use AI data lineage for audit when a selected cyber event involves AI and the company needs to connect routed traffic to approved data and business use.

Check 7: test third-party and AI route dependencies

Owner: third-party risk with the incident-response leader. Procurement and legal support the test, along with business owners and application engineering.

Pass criteria: contracts and procedures require prompt provider notice and enough incident facts for the company's materiality process. They also require continuing updates and evidence preservation, with named contacts. The company can reconcile approved AI routes with application egress and API-management telemetry. It can also reconcile those routes with provider usage. Gaps enter the incident and disclosure-control process rather than a vendor-only queue.

Evidence: provider inventory and contracts. Retain service scope and architecture records, along with the notification test and provider event sample. Add route reconciliation and corrective actions.

Place a laptop showing provider usage beside the approved-route list. Follow one endpoint to its owner and data class. Then verify its contract and incident contact, along with the response procedure. Direct traffic should appear as a discrepancy with an owner and due date.

Remediate when: a SOC report substitutes for service-specific incident terms. Embedded AI must not be absent from the dependency inventory. Provider telemetry and internal records must also use identifiers that can be joined.

Check 8: retain evidence and run an independent tabletop

Owner: internal audit or another qualified independent assurance function, with control owners responsible for remediation.

Pass criteria: the company retains source records and queries. It also retains timestamps and approvals, along with change history under approved retention rules. Reviewers retrieve an older event and verify integrity. An independent tester selects from a frozen incident population and exercises escalation and materiality support. The test also covers filing and amendment evidence, along with annual disclosure evidence.

Evidence: assurance plan and population manifest. Retain selected samples and raw outputs, along with integrity results and archive retrieval. Document findings and retest, followed by closure approval. Tamper-evident audit logs for AI covers request-event integrity; LLM audit log retention separates a configured retention period from tested retrieval.

Remediate when: the control owner chooses only successful examples, screenshots replace source exports, failed exercises lose their original evidence, or a ticket closes without independent retest.

DeepInspect

DeepInspect is a stateless proxy that sits between authenticated users or agents and LLMs, enforcing identity-bound policy on HTTP AI traffic and producing per-decision audit records. It evaluates application-supplied identity and workflow context, applies destination and content policy, inspects the response, and writes a signed, tamper-evident event with the policy version and timestamp.

Credit that capability for authenticated application-to-LLM HTTP traffic that actually traverses the proxy. The resulting records can support route reconciliation and incident triage. They can also support chronology and selected-event evidence. DeepInspect leaves identity proofing and incident classification with the registrant's assigned owners. The same applies to materiality and SEC disclosures, along with annual governance statements, provider oversight, and independent assurance. Local inference and direct browser use, plus embedded vendor AI and bypass traffic, need separate controls. Book a demo today.