Software Improvement Group launches AI code governance

Software Improvement Group has launched AI Code Governance in its Sigrid platform, a tool designed to give organisations a portfolio-wide view of AI-generated code and AI technologies.

The launch addresses a gap in software oversight as companies adopt AI coding tools without always knowing where the resulting code is used or what risks it may introduce.

The new feature lets users detect AI-generated code at both system and portfolio level. It also shows which applications use AI technologies and how those systems connect to the rest of an organisation's software estate.

Sigrid can identify AI-generated code with 95% to 99% accuracy in Java, Python and C#. The detection is intended to help engineering and IT leaders track AI use across teams and systems in real time.

The announcement comes as AI tools become more common in software development, including when developers use coding assistants outside centrally managed company accounts. Software Improvement Group argued that this has left many businesses without a clear view of where AI-written code sits in their portfolio.

According to the company, its data shows AI-generated code is more likely than human-written code to contain maintainability and security issues. That can increase technical debt over time, even when AI tools speed up early development work.

Governance focus

The product is intended to help companies govern AI use in software development at portfolio scale, rather than reviewing individual applications in isolation. It is aimed at engineering teams as well as senior IT management, which increasingly face questions about how AI affects software quality, security and long-term maintenance costs.

Luc Brandts, chief executive officer of Software Improvement Group, linked the issue to broader software quality problems already present in many organisations.

"AI amplifies what's already there. Organisations with strong software foundations will go faster and build better with AI. Organisations with technical debt, poor architecture, and ungoverned portfolios will accumulate more problems, more quickly. That was already true with AI coding assistants. With agentic AI - systems that write, test, and deploy code autonomously - the stakes are becoming even higher. The productivity gained in initial code production could be lost in future maintenance, unless you can see what's happening across your portfolio and act on it," said Luc Brandts, chief executive officer of Software Improvement Group.

That emphasis on visibility reflects a broader concern among technology leaders that AI adoption in software engineering is moving faster than the controls around it. While many companies are experimenting with generative AI, measuring the benefits remains difficult if software teams cannot track where the tools are being used and whether code quality changes as a result.

Detection method

The detection system uses stylometric analysis trained on historical enterprise code and output from current AI models. According to the company, that approach enables the platform to distinguish between code written by developers and code produced by AI systems.

Jasper Geurts, chief technology officer of Software Improvement Group, described the approach and the management questions the tool is intended to address.

"I'm proud of what our R&D teams have built. AI Code Governance detects AI-generated code across the portfolio with 95-99% accuracy. It works through stylometric analysis, trained on 25 years of pre-AI enterprise code and on what frontier models produce today. It learns the delta between them. Now, teams can see whether AI is introducing risk and fix it before it ships. Leaders can see who's using AI most, whether productivity gains are being eaten by quality and security debt, and where to focus training, tighten reviews, or adjust the definition of done. As agents start writing and deploying autonomously, AI Code Governance is part of the harness they run inside," said Jasper Geurts, chief technology officer of Software Improvement Group.

Software Improvement Group's wider business centres on software portfolio analysis and governance. Sigrid analyses more than 400 billion lines of code across more than 30,000 systems and supports more than 300 technologies.

Founded in Amsterdam in 2000, the company also has offices in New York, Copenhagen, Brussels and Frankfurt. It said it helped develop ISO/IEC 5338, a standard for AI lifecycle management.

Later versions of the product will add features to flag AI-specific security risks, measure the productivity impact of AI coding tools and provide guidance for AI-assisted modernisation, the company said.