An AI-generated incident summary lands in your inbox. It is concise, confident, and apparently ready for the board. The account is secured. Customer data is safe. Financial exposure is limited. Management can move on.
Before you forward it, ask a harder question: Does the evidence support what this summary would lead a director to believe?
That is a higher standard than checking grammar or confirming that the document sounds professional. A summary can preserve individual facts while changing their meaning. Preliminary findings become conclusions. Spending becomes total loss. A technical milestone becomes an assurance that the incident is over.
NIST’s Generative Artificial Intelligence Profile identifies confidently presented false content as a risk and warns that generated reasoning and citations can encourage misplaced trust. For board communications, the practical implication is straightforward: consequential claims need evidence, not just fluent wording.
A Clearly Labeled Hypothetical Example
The company, incident, internal notes, and AI draft below are entirely hypothetical. The regulatory references are real. The example assumes a U.S. public company subject to Form 8-K reporting. It does not describe an actual organization or personal experience.
Imagine a reviewer receives the following source packet, current as of 3 p.m. ET on November 16, 2026:
- H1 — Incident response: A compromised employee account was disabled. Investigators have not established whether other access paths exist.
- H2 — Forensics: No evidence of customer-data exfiltration has been identified in the logs reviewed so far. Coverage is incomplete, and additional logs are pending.
- H3 — Operations: The customer portal was unavailable for two hours. Service has been restored. The customer-impact assessment continues.
- H4 — Finance: External-response spending is estimated at $180,000 to date. Business-interruption losses, potential claims, and insurance recovery have not been quantified.
- H5 — Legal: Materiality remains under assessment. No conclusion has been reached about applicable disclosure obligations.
These notes establish progress, but they also preserve important unknowns. A reliable summary must do both.
The Plausible—but Unsafe—AI Draft
Illustrative AI-style draft, not an actual company statement:
The incident is fully contained, and no customer data was stolen. The brief outage has been resolved, with total losses limited to $180,000. SEC rules require disclosure within four business days of discovery. Management recommends closing the incident and returning to routine monitoring.
The draft reads like an executive summary. It is actually a series of unsupported upgrades in certainty, followed by an invented recommendation.
Compare the Claims With the Evidence
The reviewer’s job is not simply to soften the language. It is to restore the relationship between each claim and its supporting evidence.
“Fully Contained” Overstates the Response Update
H1 establishes that one compromised account was disabled. It does not establish that every access path was identified or eliminated.
Correction: The compromised employee account has been disabled; investigators have not yet established whether other access paths exist.
This distinction matters because account remediation and incident containment are not interchangeable. The board should not infer that the broader investigation is complete.
“No Customer Data Was Stolen” Erases the Evidence Boundary
H2 describes a preliminary finding within incomplete log coverage. The draft turns an absence of evidence in reviewed material into a definitive statement about what happened.
Correction: No evidence of customer-data exfiltration has been identified in the logs reviewed so far. Additional logs are pending, and the investigation remains incomplete.
The qualifiers are not clutter. They explain the scope of the finding.
“Total Losses Limited to $180,000” Changes the Financial Measure
H4 reports estimated external-response spending to date. That is neither total financial impact nor a ceiling on exposure.
Correction: Estimated external-response spending is $180,000 to date. Business-interruption losses, potential claims, and insurance recovery remain unquantified.
Similarly, H3 supports reporting that service was restored after a two-hour outage. Calling it “brief” adds a judgment while omitting that customer impact is still being assessed.
“Management Recommends Closing” Has No Source
None of the notes contains a closure recommendation. The model has crossed from summarizing evidence into manufacturing management’s position.
Correction: Remove the recommendation. If the board needs a decision, request an explicit management recommendation supported by the relevant operational, forensic, financial, and legal assessments.
The Consequential Error: The Wrong Disclosure Trigger
The draft’s SEC statement requires more than a wording adjustment. It identifies the wrong event as the trigger for the filing deadline.
Under the SEC’s cybersecurity disclosure requirements, an Item 1.05 Form 8-K is generally due within four business days after the company determines that the incident is material—not automatically within four business days of discovery. The materiality determination must be made without unreasonable delay after discovery. The distinction appears in the SEC’s announcement of the rules and the Form 8-K requirements.
Both parts belong in the correction. Replacing “discovery” with “materiality determination” without mentioning the obligation to assess materiality promptly could create a different misunderstanding: that management can postpone the assessment indefinitely.
For this hypothetical company, H5 supports neither a materiality conclusion nor a final disclosure decision. Counsel should confirm the applicable deadline, any exceptions, and other potentially applicable notification duties. Item 1.05 is not the entire universe of incident-related obligations.
The Version a Reviewer Could Forward
The following version preserves the useful compression without converting uncertainty into assurance. It remains hypothetical and is suitable for forwarding only after the responsible owners confirm H1–H5 and counsel verifies the legal language.
Hypothetical reviewed board update — November 16, 2026, 3 p.m. ET
Status: The compromised employee account has been disabled. The customer portal has been restored following a two-hour outage. Investigators have not yet established whether other access paths exist, and the customer-impact assessment continues. [H1, H3]
Data exposure: No evidence of customer-data exfiltration has been identified in the logs reviewed so far. Log coverage remains incomplete, with additional logs pending. This is not a conclusion that no data was taken. [H2]
Financial exposure: Estimated external-response spending is $180,000 to date. Business-interruption losses, potential claims, and insurance recovery remain unquantified. [H4]
Disclosure: Materiality remains under assessment; no disclosure conclusion has been reached. Item 1.05 generally requires filing within four business days after a materiality determination, which must be made without unreasonable delay after discovery. Counsel should confirm applicable obligations, deadlines, and any exceptions. [H5; SEC Form 8-K]
Proposed follow-up: Request a further update within 24 hours, or sooner if findings materially change, covering containment, data exposure, financial impact, and the disclosure assessment.
Board action: This update does not request incident closure. Any approval request should be presented separately with management’s recommendation and supporting evidence.
Notice what changed: the revised update separates established facts, unresolved questions, regulatory requirements, and proposed next steps. The follow-up is clearly labeled as a proposal—not presented as an existing management commitment.
A Practical Checklist Before You Forward
Make the review repeatable rather than dependent on someone noticing a suspicious sentence.
- Confirm the reporting cutoff. A statement can be accurate for an earlier update and misleading when forwarded later.
- Trace consequential claims. Connect each decision-relevant assertion to a dated source and responsible owner. Source labels such as H1 are useful only if reviewers can access the underlying material.
- Preserve uncertainty. Keep qualifiers such as “to date,” “reviewed logs,” and “not yet determined” when they change the meaning.
- Check numbers and scope. Verify the amount, category, period, and completeness. Spending is not total loss; restored service is not completed containment.
- Verify legal statements independently. Open the authoritative source. Do not treat a model-generated citation or paraphrase as verification.
- Separate findings from recommendations. Confirm who is proposing an action and whether that person has actually approved the recommendation.
- Retain the review record. Preserve the source packet, draft, corrections, and approval record under the organization’s applicable retention and handling requirements.
This workflow reflects NIST’s emphasis on distinguishing facts from inference, acknowledging uncertainty, and making information traceable to evidence. It also focuses human attention where an error could change a decision.
Forward Evidence, Not Confidence
AI can help compress a complicated incident into a readable board update. It cannot make incomplete evidence complete, resolve materiality by implication, or speak for management without authorization.
Before forwarding your next AI summary, identify every claim that could change a director’s decision and check it against the source. Correct unsupported certainty, preserve unresolved questions, and make proposed actions explicit.
The forwarding standard is simple: concise enough for directors to use, explicit enough that uncertainty cannot be mistaken for assurance.