Systems that maintain themselves: measure, correct, update, approve
Published
The short answer
How does software stay secure and current after the handover without unreviewed changes going live?
By being built from the start to measure continuously whether it does what it should, to detect deviations and newly known security flaws itself, and to prepare and test corrections and updates automatically. A change only goes live once you have approved it, with traceable evidence.
From the handover on, a system ages
Most software projects end with a handover. From then on the system ages: dependencies acquire security flaws, measurements drift, and faults only surface when a customer reports them.
Self-maintenance does not mean a system works on itself unsupervised. It means the work of maintenance is prepared before a person has to start it, and the decision stays with you.
Measure
Health checks, telemetry and evaluation show at all times whether the system does what it should. Latency, errors and cost are measured, and for an assistant the quality of its answers too: are they backed by sources, and does it decline where it should? An open standard such as OpenTelemetry keeps the measurements independent of any single tool.
Correct
When a measurement drifts or a scan reports a security finding, a controlled correction starts. It passes the same tests and security checks as any other change. If anything fails, the system returns to its last good state.
That requires tests that really check. A test suite can be entirely green while single tests check nothing. Mutation testing breaks the code on purpose and shows whether the responsible test goes red.
Stay current
Dependencies and containers are checked continuously. The basis is a software bill of materials (SBOM), for example in the CycloneDX format, which records which components are included in which version. Newly known vulnerabilities are matched against it; container and secret scanning complete the picture.
When a flaw becomes known, the update is prepared and tested automatically. Whether it is applied is your decision.
Approve
Every change waits for your approval with traceable evidence: what changes, which checks passed and how the change is rolled back. Prepared automatically, approved by you.
The approval is not a formality. It keeps responsibility where it belongs and makes every change traceable afterwards.
The framework behind it
This approach is in line with recognised guidance for secure software development, such as NIST's Secure Software Development Framework (SP 800-218), and with the expectation that vulnerabilities are handled over the whole lifecycle of a product. For products with digital elements, the EU Cyber Resilience Act (Regulation (EU) 2024/2847) will make this an explicit requirement.
Sources and evidence
Planning something along these lines? Briefly describe your project and you will get an honest assessment.
More articles
All articles- Why permissions belong in front of the model, not in the prompt
Is it enough to tell an AI assistant in its prompt not to reveal confidential content?
- AI in your own data centre or an EU cloud: how to keep control of your data
Can a company use AI without its data leaving the building or the EU?
- Knowledge search with cited sources: cite or decline
What should an AI knowledge search do when its documents hold no answer?