AI Coding: The Dark Side of 97% Adoption in Enterprises

AI Coding: The Dark Side of 97% Adoption in Enterprises

In the world of software development, AI-powered coding tools are no longer the future; they are the present. According to Black Duck’s “State of AI-Powered Software Development” report, the adoption of such tools in large enterprises (with at least 500 employees) has reached 97 percent. Developers save an average of eight hours per week using them, which on paper promises staggering productivity gains. However, beneath the surface lie serious problems that can undermine these benefits.

The survey, with 78 percent of respondents from the technology and SaaS sectors and 52 percent from C-level or senior management, paints a clear picture: technology adoption has far outpaced the establishment of regulation and oversight. This gap is not just a theoretical problem but a tangible business and security risk.

Do you have a question about AI security? You can reach us here:

The Productivity Paradox: The Shifting Bottleneck

While AI does indeed accelerate code writing, the report highlights that the bottleneck has merely shifted to a later stage in the process. Ninety percent of respondents experience some form of problem with AI-generated code. In the development lifecycle, the following areas have become the most time-consuming:

  • Manual code review (52 percent)
  • Security testing (51 percent)
  • Refactoring the generated code (48 percent)

In a corporate context, this means that the initial time savings are consumed by the quality assurance and correction of the generated code. From an AIQ standpoint, this phenomenon perfectly illustrates that tools alone are not the solution; they must be viewed as part of an integrated, security-conscious development process (DevSecOps), not as a silver bullet.

Hidden Vulnerabilities and Corporate Responsibility

Even more serious than the efficiency issues is the security risk. According to the report, nearly half of all AI-generated code contains vulnerabilities. This is a shocking figure that exposes companies to significant danger. 64 percent of respondents are moderately or extremely concerned that AI-generated code will introduce security defects or vulnerabilities into their systems.

From AIQ’s perspective, this issue is directly linked to several points in the OWASP LLM Top 10. Although the coding assistant itself may not be vulnerable, the code it generates can cause “Insecure Output Handling” issues, which are a breeding ground for classic OWASP Top 10 vulnerabilities. If a model was trained on insecure code snippets, it could also relate to the OWASP LLM Top 10’s “Training Data Poisoning” point, leading to hidden, hard-to-detect flaws.

“AI coding assistants have permanently changed the economics of software development, and the productivity numbers make that undeniable. But the data also clearly shows that speed without governance is a liability, not an advantage.” – Jason Schmitt, CEO of Black Duck.

The Key to the Solution: Governance and Automation

The most important takeaway from the report is that a proper governance framework is essential to realizing the true benefits. Currently, only about a third of teams have full governance over their AI coding tools. However, where it exists, the results speak for themselves: these teams are 55 percent more likely to report significant efficiency gains.

Developers and leaders have recognized the root of the problem. 84 percent of respondents still prefer human oversight, and 68 percent consider an automated system for tracking the deployment of AI-generated code to be extremely important. Furthermore, 56 percent would like a dedicated AI security agent separate from the coding tool itself.

The AIQ Perspective: EU AI Act and the Corporate Market

From an AIQ standpoint, the findings of the Black Duck report are particularly relevant to the European corporate sphere. The approaching implementation of the EU AI Act places the issues of accountability and transparency at the forefront. While a coding assistant itself may not be classified as a high-risk AI system, the software it helps create—for example, an application used in finance, healthcare, or critical infrastructure—easily could be. In the event of a data breach or system failure resulting from an AI-generated vulnerability, the responsibility will lie with the company that developed and operates the software.

The GDPR principle of “privacy by design and by default” also requires that every element of the development process, including AI tools, be under strict control. During an audit, a company must be able to prove that it has done everything possible to minimize vulnerabilities. The lack of governance and documented, automated controls poses a serious compliance risk.

The lesson is clear: implementing AI coding tools is not a simple IT procurement but a strategic decision that requires profound changes in development culture and security processes. The solution is not to ban these tools, but to build a robust, automated governance and auditing framework that includes LLM red teaming and continuous security validation.

Attila Rácz-Akácosi

Independent AI Security Specialist

Two decades of analytical and systems-oriented experience. I have been working with artificial intelligence since 2017. In recent years, I have specialized in AI/LLM security and AI Red Teaming. Systems-level thinking instead of endless vulnerability checklists.