Harness Unveils Alarming AI Visibility Crisis Impacting Enterprises

In a world increasingly driven by artificial intelligence, organizations find themselves in peril as they sprint to integrate AI into every facet of their operations. A recent report from Harness, the AI DevOps Platform™, highlights a troubling trend: many enterprises are losing visibility into the applications and components powered by AI, leading to a growing unseen risk landscape.

The newly released report, The State of AI-Native Application Security 2025, paints a stark picture of how AI, already rapidly evolving, is becoming a double-edged sword in enterprise settings. As companies adopt AI technologies at an unprecedented rate, security teams report an alarming rise in incidents related to AI implementation. According to the study, a staggering 75% of security professionals identify shadow AI as a bigger threat than the shadow IT issues faced in previous years. Shadow AI refers to the unmonitored use of AI technologies without proper oversight, creating vulnerabilities within an organization.

The report's findings are based on a survey involving 500 security experts and decision-makers across the USA, UK, France, and Germany, revealing widespread concern among professionals. Key insights from the research include the fact that 62% of respondents lack visibility into where large language models (LLMs) are being utilized in their organizations. Adding to the complexity, 74% predict that the risks associated with AI will far exceed those posed by API sprawl.

Moreover, as cybercriminals become increasingly sophisticated, 82% of the respondents believe that AI-native applications represent a prime target for attacks. Notably, incidents of LLM prompt injection, vulnerable code, and jailbreaking have already been reported by 76%, 66%, and 65% of survey participants, respectively. This highlights a clear message: data and systems accessible through AI technology are now under constant threat.

One of the alarming statistics is that 62% of developers do not take responsibility for securing these AI-native applications, while only 43% design them with security in mind from the outset. Adam Arellano, Field CTO at Harness, underscored this gap by stating, "Shadow AI has become the new enterprise blind spot, and traditional security mechanisms cannot keep pace with the rapidly changing nature of adaptive AI models."

Unfortunately, this issue isn't merely confined to technical challenges. It reveals a deeper division in how security is perceived within organizations. The survey indicates that 75% of security leaders feel that developers often view security measures as an obstruction to innovation. Furthermore, 62% of developers report they are pressed for time and often lack the necessary training to adequately secure AI applications, leading to a significant deficiency in AI-native security practices.

In an age where AI is becoming the backbone of numerous applications, it is imperative for organizations to prioritize security in the early stages of development. According to Arellano, security must be considered in all phases of the software lifecycle — before, during, and after implementation — to maintain effective oversight and control.

To address the visibility crisis, Harness offers several recommendations for organizations aiming to boost their AI-native security resilience:
1. Embed Security from the Start: Establish a shared governance framework between security and development teams to ensure consistent oversight.
2. Discover AI Components: Proactively identify and monitor incoming AI components as they emerge within the organization.
3. Real-Time Monitoring: Implement systems that provide real-time visibility into AI components and model outputs to quickly detect anomalies.
4. Dynamic Testing: Conduct dynamic tests to assess applications against AI-specific threats before launching them into production.
5. Protect Production Environments: Focus on safeguarding AI-native applications post-deployment to minimize the risks associated with sensitive data exploitation.

In closing, as enterprises navigate this new wave of artificial intelligence, they must confront the challenge of maintaining security amid complexity. With AI's rapid integration into workflows, the call for characterized governance and collective responsibility never echoed louder. For comprehensive insights, organizations can explore the complete report by Harness detailing their findings on AI-native application security. Read the full report here.

About the Research: This report stems from a survey conducted by Harness with 500 security practitioners and decision-makers tasked with securing AI-native applications, facilitated by Sapio Research. The demographic includes 200 respondents from the USA, and 100 respondents each from the UK, Germany, and France.

About Harness: Harness, as the AI DevOps Platform™, is committed to accelerating the software delivery lifecycle for engineering teams, leveraging automation at every stage to reduce manual workloads and enhance productivity. Renowned companies like United Airlines and Citibank trust Harness to cut costs and increase efficiencies across their DevOps practices.