AI is becoming deeply embedded in enterprise systems—often without explicit decisions or oversight from leadership. Haris Shamsi, founder of Yottabyte, highlights that AI is entering the workplace not through centralized deployment but as bundled features within existing tools. From writing assistants and customer support bots to analytics platforms, AI is increasingly adopted by teams seeking quick automation, not realizing the risks that come with it. This quiet rollout, often initiated by vendors or user teams, has led to what cybersecurity professionals are calling “unconscious vendor creep.”
At RSAC 2025, Jon France, CISO at ISC2, described this growing trend where AI appears unexpectedly across business applications. Whether it’s a new button in a CRM or an added field in a ticketing tool, AI is being deployed without architectural review, threat modeling, or input from security professionals. These implementations can introduce significant risk, especially when teams assume that embedded AI is secure simply because it comes from a trusted vendor or has passed procurement procedures. However, many AI features are not secure by default and lack proper transparency.
A major concern lies in the lack of understanding around how these systems handle sensitive data. For example, if an AI-powered writing assistant draws from internal documents or customer campaigns, it may unintentionally leak confidential information. Similarly, AI in customer support tools may retain ticket content, store it externally, or use it to train future models. Without clarity on how data is processed, logged, or shared, businesses may be exposing themselves to privacy and compliance violations.
This phenomenon is contributing to the emergence of “shadow AI,” mirroring the earlier rise of shadow IT. Teams are integrating AI-powered tools like resume scanners, content generators, or research bots without security clearance. If these tools interact with internal systems or client data, they create potential exposure, especially if they retain prompts, log interactions, or retrain on user input. In 2024, IBM reported that over half of AI-related breaches resulted from misconfigurations or a lack of visibility—not complex technical flaws, but governance failures stemming from insufficient oversight.
With AI now built into standard workplace software, older policies like blanket bans on tools such as ChatGPT are no longer practical. Enterprises are realizing the need for updated AI policies that define where, how, and under what constraints AI can be used. It’s essential to document which tools contain AI, what data they access, and who approves their usage. Organizations must enforce data boundaries, restricting sensitive information from being processed or stored by external models, particularly those operated by third parties.
Security teams are being urged to treat AI systems as dynamic and high-risk, given their ability to evolve, adapt, and act unpredictably. Unlike traditional software, AI’s logic is not always consistent or transparent, making its behavior difficult to validate. As regulatory landscapes evolve globally—such as the European Union’s AI Act or US executive orders on innovation and national security—businesses will need flexible and well-documented AI governance frameworks to remain compliant across jurisdictions.
Raising awareness among employees is now a key part of cybersecurity strategy. Just as phishing training became essential in earlier digital eras, AI literacy must become standard. Staff should understand where AI is embedded in their workflows, how it interacts with company data, and what warning signs to look for. Without this awareness, AI can quietly become a new attack surface—one that acts fast, learns quickly, and can cause significant damage if left unchecked.
