Software vulnerabilities do not arise according to a schedule.The Chief Security Officer of Deutsche Telekom has commented on this topic and in doing so corrected widespread ideas about artificial intelligence.His key point: code analysis with AI models is not an occasional special run, but a task that must run around the clock.He sees a particular risk in open source.
The background is directly relevant for operators of web applications and content management systems.Zero-day vulnerabilities are weaknesses for which no patch exists at the time they are exploited.Anyone who only reacts when a vendor publishes an update is, by definition, too late within this window.
Why continuous operation and not sampling
Classic security checks often follow a project rhythm: a penetration test before launch, an audit before relaunch, little in between.This pattern no longer fits applications that are continuously developed further and whose dependencies change weekly.
The argument from the Telekom environment amounts to providing analysis capacity permanently instead of selectively.In practice, this means: checks run on every commit, on every build and on every update of an external library – not once per quarter.
For mid-sized organisations, this is initially an unpleasant message because it sounds like additional effort.In fact, the effort tends to shift rather than grow: away from elaborate individual checks, towards automated steps in the build chain, that is, in the process that turns source code into a runnable system.
Open source as a particular problem area
The fact that the security officer highlights open source of all things is not an argument against open-source software.It is an indication of a structural characteristic of modern projects: the majority of shipped code usually does not come from one's own house, but from third-party packages, frameworks and libraries.
In a typical web project, that quickly means hundreds of dependencies, many of which in turn bring their own dependencies.This transitive depth is the actual point: a team can know the part it wrote itself very precisely and still be unable to say what is integrated three levels deeper.
Added to this is observability.Publicly visible code is equally readable for defenders and attackers.Anyone searching automatically for patterns can do so across a very large number of projects at the same time – which makes analysis scalable on both sides.
What AI models can achieve in code analysis
Static code analysis has existed for decades.Its well-known problem is false positives: tools report more anomalies than a team can sensibly work through, and at some point the reports are ignored.
The added value of models lies above all in incorporating context – for example, whether a potentially dangerous function even comes into contact with user input.This changes prioritisation less than detection.What is decisive is not generating as many findings as possible, but sorting the few genuinely exploitable ones to the top.
At the same time, the limitation that also underlies the clearing up of AI myths applies: models do not replace security architecture.They do not find errors in business logic that is not stored anywhere as a rule, and they do not decide which risk a company wants to bear.
Transfer to CMS and TYPO3 projects
A TYPO3 project consists of the core, extensions of various origins and individually developed code.It is precisely this mixture that makes the described problem concrete: the attack surface changes with every integrated extension, even without development work of one's own.
Sensible steps can be implemented without a large apparatus:
- Maintain an up-to-date directory of all components used, including version status, also for transitive dependencies.
- Hook dependency checks into the build chain automatically instead of triggering them manually.
- Determine in advance who assesses findings and within what deadline critical reports are handled.
- Treat update capability as a project goal: anyone who heavily overrides the core or extensions extends their reaction time with every patch.
- Select extensions according to maintenance status – an unmaintained package is a permanent risk, regardless of its range of functions.
Organisation beats tool
The obvious fallacy is that one merely needs to purchase another analysis tool.In fact, the benefit only arises where findings are reliably processed.A report at three in the morning is of little help if someone is only responsible three weeks later.
Therefore, around-the-clock analysis necessarily includes responsibilities, escalation paths and a realistic assessment of how quickly an update can be deployed.This part cannot be automated, but it shapes the actual security situation more strongly than the choice of scanner.
For web and software projects, this means: security is finally shifting from project completion into ongoing operation and should be reflected accordingly in the budget.Anyone who documents dependencies, maintains update capability and combines automated checks with clear responsibilities noticeably reduces the time window between a vulnerability becoming known and being closed.
AI-supported analysis primarily helps with prioritisation – the decision on how quickly to respond remains an organisational one.