A compliance tool must not present a deadline as a legal obligation without being able to justify its source, its scope and its starting point.
On 10 September, comparing the deadlines produced by our regulatory reporting engine with the applicable texts, we identified an error in our model of NERC CIP-008-6: our software showed a final report within 30 days, while that requirement provides for no such milestone. The deadline came from an analogy with other regimes. It was consistent with some incident reporting practices, but it did not rest on the text we claimed to represent.
1. The finding: a deadline with no basis in the text
Our engine showed three milestones for NERC CIP-008-6: initial notification within one hour, update within seven days, final report within thirty days. Comparison with the text showed that the third was not part of the requirement concerned.
The problem was not a badly entered duration. The software presented as an obligation drawn from a text a deadline that came from an extrapolation of our own model. In a compliance tool the distinction is central: an internal deadline can be useful for organising work, but it must not be presented as a legal requirement if its source does not justify it.
2. What CIP-008-6 actually provides
Requirement R4 distinguishes several situations, for the systems and equipment it covers:
- a reportable cyber security incident is the subject of an initial notification within one hour of being determined as such;
- an attempt to compromise the systems covered is notified by the end of the calendar day following the corresponding determination;
- updates, where there are any, are provided within seven calendar days of determining new or changed information.
These deadlines are not interchangeable: they depend on the nature of the event and, for updates, on new information appearing. The standard does not provide, in that requirement, for a systematic final report within thirty days.
Scope matters too: CIP-008-6 concerns the entities and electric systems falling under the applicable North American framework. It is not a universal obligation. Primary source: NERC CIP-008-6, Cyber Security — Incident Reporting and Response Planning.
3. How the error got past our checks
The wrong deadline caused neither a software exception nor a test failure. The module produced exactly the value its own tests expected.
That was the limit of our validation: we checked that the software applied its configuration, not that the configuration matched the text it was meant to represent. A shorter deadline may look prudent; it is not for that reason a faithful representation of the applicable law, and it makes it harder to tell a legal requirement from an internal precaution.
4. The correction: tying every rule to its source
The thirty-day milestone was removed. The data model was strengthened: a framework declared verified carries a precise regulatory reference, text and article included, and a test refuses a framework marked verified without a cited source. A second test checks that a framework explicitly states where the engine simplifies the text.
That control is a traceability guarantee, not an interpretation one: a reference can be valid and still be misread, incomplete, or inapplicable to a given organisation. Each framework therefore documents its source, its scope, its trigger, its milestones and its caveats.
One point remains to improve: recording the date of the last comparison with the text, and the version of the source consulted. Without it, it is hard to show when a rule was verified, and whether that verification still holds after a regulatory change.
5. Four frameworks, four distinct scopes
The engine covers four reporting frameworks. They do not reduce to a common timeline.
- NIS2, Directive (EU) 2022/2555, art. 23(4): early warning, incident notification, final report. The text runs the final report from the notification; our engine runs it from the moment of becoming aware.
- DORA, Delegated Regulation (EU) 2025/301, art. 5: initial notification, intermediate report, final report, with chained starting points. The text also requires an initial notification within four hours of classification as a major incident: that clock is not modelled.
- NERC CIP-008-6, R4: the deadlines vary depending on whether it is a reportable incident, an attempt to compromise, or an update.
- Regulation (EU) 2024/1689 on AI, art. 73: reporting a serious incident involving a high-risk system. Two accelerated cases are not modelled: two days for a widespread infringement, ten days in the event of a death.
Their presence in the same software means neither that they target the same organisations, nor that they apply simultaneously to a given user.
6. A calculated deadline is not necessarily a legal one
The verification revealed a structural limit of our model: it has a single clock, the moment of becoming aware of the incident, where several regimes chain deadlines triggered by different events.
DORA illustrates it. The intermediate report is expected within 72 hours of the initial notification, and the final report at the latest one month after the intermediate one or its last update. Calculating those milestones from the moment of becoming aware alone simplifies the model: the deadlines shown fall earlier than those of the text, therefore never late, but they do not reproduce the chain of triggers.
We therefore separate two notions. The regulatory deadline is calculated from the triggers and conditions of the text. The operational deadline is an internal target, possibly stricter, adopted to keep a margin. Where the engine does not model every step of a regime, the corresponding caveat is displayed next to the deadline.
7. What we still have to improve
- record the verification date and the version of the text consulted;
- plan a review when a text changes;
- test the triggers and the chaining of deadlines, not only the durations;
- explicitly identify conditions, exceptions and cases not modelled;
- have the interpretations reviewed by a legal professional.
Automated tests detect a missing reference, an inconsistent rule or a regression. They do not demonstrate that a legal interpretation is correct.
8. Why we publish this
We build a tool meant to help organisations understand their obligations and organise their incident response. Every deadline must be tied to a source, a scope, a triggering event and explicit assumptions. Where the model simplifies a rule, the simplification must be visible; where a verification is missing, the software must not suggest it was done.
This analysis was carried out by an engineer, from the texts and the implementation. It does not replace a formal legal validation, which remains to be done to confirm the interpretation of the various regimes and their conditions of application. We prefer to document that limit rather than present a technical verification as a legal guarantee.