The rapid adoption of Artificial Intelligence (AI) tools in the enterprise is creating a new set of complex challenges for organizations. Based on a recent discussion we had with a company expanding its AI usage, several core problems have emerged, primarily centering on governance, security, and compliance.
Here are the critical problems companies are facing today as they move beyond initial, limited AI use:
1. The Critical Gap in Governance and Security Guardrails
The most immediate and pressing concern for many companies is the lack of proper oversight. While AI adoption is often driven by a desire for efficiency, such as leveraging tools like Claude for coding or building customer service agents, this rapid deployment often occurs without necessary security infrastructure.
- Shadow AI and Risk: Developers and employees are frequently using enterprise-grade AI tools (e.g., enterprise ChatGPT licenses) without any effective guardrails, allowing them to “do almost anything” and posing significant security risks.
- Need for a Framework: There is a struggle to implement a comprehensive governance framework—a system to “framework, map, monitor, and manage” all AI tool usage—to prevent complications and manage risk.
2. Managing a Fragmented AI Tool Landscape (Tool Sprawl)
The current market is characterized by a mix of platforms, which leads to complexity in license management and user dissatisfaction.
- Multiple Licenses: Organizations often juggle various tool licenses—including Copilot, Claude, and multiple ChatGPT seats—creating an inconsistent and difficult-to-manage environment.
- Outdated Internal Solutions: In some cases, internally developed, older chatbots (like those based on GPT-3 or GPT-4) fail to meet the performance and feature expectations set by newer, public-facing models, leading to internal resistance and a push for more advanced, yet unsecured, tools.
3. Choosing and Implementing a Compliance Standard
As AI risk becomes a top-tier security issue, companies are forced to commit to a formal standard, but the choice and implementation are challenging.
- The Compliance Divide: For companies that do not operate internationally and are not ISO compliant, the choice often falls to domestic standards. Many are rejecting the global ISO AI risk management 42001 in favor of frameworks like the NIST AI Risk Management Framework to achieve security compliance goals similar to NIST CSF (Cybersecurity Framework).
4. The Rising Threat of AI-Specific Attacks
The risk is not just about unmanaged data exposure; it is about active, malicious attacks targeting the AI models themselves.
- EDR for AI: Companies require security solutions that function as an “EDR (Endpoint Detection and Response) for AI” to focus on preventing attacks against both internal and external chatbots.
- Preventing Malicious Use: Organizations must be able to collect telemetry to establish a baseline for governance, but, more critically, they must actively block malicious uses of AI agents, which include:
- Attacks to “jailbreak” AI models.
- Attempts to extract confidential training data from proprietary models.
- Exploiting connections to back-end systems (like customer databases) to cause data leaks or unauthorized actions.
In summary, for companies to successfully scale AI, the focus must shift immediately from simply deploying tools for efficiency to establishing robust security and governance layers that can consistently map, monitor, and manage the growing complexity and risk landscape.