On 23 October 2026, three providers presented innovations almost simultaneously that address the same problem area: securing the software supply chain, i.e. all the components that come together on the path from source code to finished deployment. Amazon Web Services, GitHub and JFrog start at different points, but pursue a common direction: checks should be automated and take place as early as possible in the development process.

AWS Continuum: detect and remediate in one step

Amazon Web Services has made AI-powered detection and remediation of code vulnerabilities generally available in AWS Continuum. What is remarkable is less the detection – static code analysis has been part of the standard repertoire for years – than the coupling with a remediation suggestion. The tool therefore does not stop at a list of findings, but supplies a correction suggestion with it.

For teams, this shifts the workload. The effort is no longer primarily in finding, but in assessing: Is the suggested fix technically correct, is it compatible with the existing architecture, does it cover the actual attack path? This review cannot be delegated; it remains the task of experienced developers.

GitHub Package Firewall: control at the front door

GitHub has released the Package Firewall as a public preview for Enterprise Cloud customers. It checks package requests and downloads and is intended to block malicious npm packages before they even land in a project. npm is the package manager of the JavaScript ecosystem and has been a preferred target for attacks via manipulated or similarly named packages for years.

The approach differs fundamentally from downstream checking: instead of scanning installed dependencies after the fact, the control takes effect at the point of retrieval. This shortens the window in which a compromised package can become active on developer machines or in build environments. The fact that the function is initially available as a public preview and only to a specific group of customers should be taken into account when planning – preview versions are generally not intended for production-critical dependencies.

JFrog: shadow AI on developer machines

JFrog is expanding shadow AI detection in its AI Security suite. Shadow AI refers to AI tools and models that are used in workflows without approval or knowledge of IT managers. New is BYOK support for AI Asset Scanning. BYOK stands for "Bring Your Own Key" and means that customers can bring their own keys for encryption instead of using only the provider's.

This announcement addresses a topic that is underestimated in many organisations. Locally installed models, unofficial extensions in development environments and services that send code out for processing often arise for pragmatic reasons and without malicious intent. For security considerations, they are nevertheless relevant because they create dependencies and data flows that do not appear in any inventory list.

Three tools, three levels

The announcements complement each other without the providers having coordinated this. They each cover a different level of the supply chain:

  • Own code: AWS Continuum addresses the vulnerabilities that arise in self-written or AI-generated code.
  • Third-party dependencies: The GitHub Package Firewall controls what enters the project from public package sources.
  • Tools and environment: JFrog takes the developer machines themselves into view, in particular unapproved AI components.

Caution is appropriate regarding the sources: the available announcements cite neither detection rates nor prices, licensing models or concrete technical details on how they work. Anyone who wants to evaluate one of the tools must obtain this information from the respective provider. The level of maturity also differs: AWS speaks of general availability, GitHub of a public preview.

Why now of all times

The timing coincidence is no accident. If a growing share of code is created with AI support, the volume of changes grows faster than the capacity for manual reviews. At the same time, language models occasionally generate references to packages that do not exist at all – a gap that attackers can exploit by deliberately creating such package names. Both developments increase the pressure to automate controls and move them closer to the point of origin.

What project managers can derive from this

Regardless of whether one of these specific products is an option, some consequences can be noted:

  1. Inventory dependencies. Without a reliable overview of which packages are in use in which versions, all downstream measures come to nothing.
  2. Formulate a policy for AI tools. Whoever regulates which assistance systems and models are permitted in development and which data they may process reduces shadow AI before detection tools become necessary.
  3. Do not adopt automatic fixes without review. A remediation suggestion is a suggestion. The review remains mandatory, even if it changes as a result.
  4. Shift controls into the pipeline. Checks that run only locally or only occasionally are skipped in day-to-day project work.Anchored in the build pipeline, they work reliably.

For web and software projects, this development means above all one thing: security control is shifting from a one-off check before release to an ongoing function of the development environment.Anyone who uses AI support in the team should establish the corresponding checking mechanisms at the same time and not treat them as a later addition.The practical entry point is less a product decision than an organisational one: clear rules for dependencies and tools, documented in a traceable way and enforced in the pipeline.

Sources