New Harness Report Highlights the Risks of AI Agents Without Proper Controls
The recently released report titled The State of Agent DLC 2026 by Harness sheds light on a crucial gap in the confidence enterprises have in AI agents compared to the actual security and governance measures in place. As companies increasingly adopt AI technologies, the report highlights a stark reality: many organizations are overly confident in their AI systems while lacking the necessary controls to ensure their safe and effective operation.
Conducted with responses from 700 engineering leaders, the survey reveals that while enterprise confidence in AI agents is high, the corresponding controls for testing, securing, and governing these agents fall dramatically short. A significant 74% of respondents expressed confidence in their ability to detect production-impacting failures, yet only 19% have the proper mechanisms to prevent faulty releases, such as automated gates.
The report points out that AI agents behave in unpredictable ways, diverging significantly from traditional deterministic software, which makes their governance more complex. For example, organizations often rely on controls designed for predictable software, which do not hold up against the inherent variability of AI agents. This inconsistency has already led to increased incidents of security breaches and budget overruns in many firms.
Several key findings from the survey illustrate this confidence-control gap:
The report emphasizes the need for organizations to treat the AI agent lifecycle as a distinct discipline with specialized practices for evaluation, testing, security, and deployment, rather than relying on models used for traditional software. Harness recommends implementing dedicated tools and progressive rollouts to manage AI behavior changes more effectively.
Keith Mann, Field CTO at Harness, commented on the findings, stating that the pattern of overconfidence without proper governance is not new in the tech landscape, having previously been observed with cloud and mobile technologies. “However, AI agents differ in their operation, necessitating a reevaluation of how businesses manage their implementations and monitor their performance,” he noted.
Organizations that manage to close this confidence gap are likely to emerge as leaders in the development and deployment of AI technologies. They are already adopting more rigorous frameworks dedicated to the unique challenges posed by AI agents, such as implementing canary and blue/green deployments, rather than continuing to integrate these changes into conventional software pipelines.
In a marketplace where speed and reliability are crucial, the findings serve as a wake-up call for companies embracing AI agents. It is imperative that they not only focus on deploying these systems but also prioritize the establishment of proper controls that ensure their ongoing performance, security, and overall effectiveness. To delve deeper into these insights and recommendations, access the full State of Agent DLC 2026 report at Harness’s official website.
Conducted with responses from 700 engineering leaders, the survey reveals that while enterprise confidence in AI agents is high, the corresponding controls for testing, securing, and governing these agents fall dramatically short. A significant 74% of respondents expressed confidence in their ability to detect production-impacting failures, yet only 19% have the proper mechanisms to prevent faulty releases, such as automated gates.
The report points out that AI agents behave in unpredictable ways, diverging significantly from traditional deterministic software, which makes their governance more complex. For example, organizations often rely on controls designed for predictable software, which do not hold up against the inherent variability of AI agents. This inconsistency has already led to increased incidents of security breaches and budget overruns in many firms.
Several key findings from the survey illustrate this confidence-control gap:
- - Visibility Issues: While 77% of organizations believe they have complete visibility over their AI agents, only 44% utilize active tools to verify this claim. This lack of active monitoring can lead to unforeseen issues, as organizations struggle to maintain an accurate inventory of AI agents at play.
- - Inadequate Testing Mechanisms: A large majority of respondents, 74%, feel assured that their testing processes will catch errors before production, but only 19% have actual systems in place that would block harmful updates from proceeding.
- - Limited Response Capabilities: When an AI agent behaves unpredictably, 76% of leaders believe they could disable it within 15 minutes. However, barely a third have an immediate kill switch available, highlighting a critical vulnerability.
- - False Sense of Security: Although 75% of organizations claim their AI systems are secure end-to-end, the rate of security incidents among this group is nearly identical to that of the overall population, suggesting a disconnect between confidence and actual security measures.
The report emphasizes the need for organizations to treat the AI agent lifecycle as a distinct discipline with specialized practices for evaluation, testing, security, and deployment, rather than relying on models used for traditional software. Harness recommends implementing dedicated tools and progressive rollouts to manage AI behavior changes more effectively.
Keith Mann, Field CTO at Harness, commented on the findings, stating that the pattern of overconfidence without proper governance is not new in the tech landscape, having previously been observed with cloud and mobile technologies. “However, AI agents differ in their operation, necessitating a reevaluation of how businesses manage their implementations and monitor their performance,” he noted.
Organizations that manage to close this confidence gap are likely to emerge as leaders in the development and deployment of AI technologies. They are already adopting more rigorous frameworks dedicated to the unique challenges posed by AI agents, such as implementing canary and blue/green deployments, rather than continuing to integrate these changes into conventional software pipelines.
In a marketplace where speed and reliability are crucial, the findings serve as a wake-up call for companies embracing AI agents. It is imperative that they not only focus on deploying these systems but also prioritize the establishment of proper controls that ensure their ongoing performance, security, and overall effectiveness. To delve deeper into these insights and recommendations, access the full State of Agent DLC 2026 report at Harness’s official website.
Topics Business Technology）
【About Using Articles】
You can freely use the title and article content by linking to the page where the article is posted.
※ Images cannot be used.
【About Links】
Links are free to use.