Your EDR is blind to Python: Here’s what’s already running in your endpoints

The proliferation of Generative AI (GenAI) and Large Language Models (LLMs) has ushered in a new era of innovation, but it has simultaneously created a formidable, often-unseen security challenge: Shadow AI.

This phenomenon is rapidly transforming the corporate technology landscape, positioning Python (the “lingua franca of AI”) as the new attack surface and the modern equivalent of the problematic Excel macro.

Why Traditional Controls Fall Short Against Shadow AI

Traditional endpoint security and governance solutions were not designed for the current reality. The speed and ease with which business users can now generate and deploy Python scripts (often for beneficial, innovative purposes) means that established controls are fundamentally inadequate.

The data is clear and alarming:

• 20% of organizations have already been breached via Shadow AI, yet 63% have no formal AI governance policy (IBM Cost of a Data Breach Report, December 2025).
• Gartner projects that 40% of organizations will face a Shadow AI breach by 2030.
• 69% of organizations already suspect their employees are using prohibited GenAI tools.

The core issue is that employees, empowered by tools like GPTs and AI co-pilots, are automating processes and building small applications (often in Python) that directly handle sensitive company data.

For someone who has never coded before, simply asking an LLM to “track a USPS package” or “extract data from a spreadsheet” results in immediate, executable Python code.

The risks are no longer theoretical; they are manifesting across all regulated industries, creating severe security and operational risks:

The Six Critical Risk Vectors Your Traditional Controls Miss

When unsupervised Python scripts run on endpoints, they introduce a set of risks that standard Endpoint Detection and Response (EDR), Data Loss Prevention (DLP), and Mobile Device Management (MDM) tools are blind to:

1. Hardcoded Credentials: Scripts frequently contain embedded usernames, passwords, or API keys, creating easily exploitable security vulnerabilities.
2. Anomalous Activity Outside of Business Hours: Automated scripts often run at odd hours, making them difficult to distinguish from malicious activity.
3. Vulnerabilities and Malicious Packets: LLMs can recommend outdated or vulnerable Python libraries, and these scripts can initiate network connections that carry malicious payloads.
4. Database Queries: Scripts can establish direct connections to company databases, bypassing application-level security controls.
5. Reading, Manipulating, and Exfiltrating Files: Scripts are designed to process data. This includes reading sensitive files and exfiltrating them outside the corporate environment (e.g., to personal cloud storage or via external APIs).
6. Script Loss and Operational Risk: If the employee who created the script leaves the company, valuable, functioning automations (the good scripts) can be lost, leading to critical operational failures.

The ultimate question for the board is: “How many Python scripts are processing customer data on endpoints right now, and who approved them?”

While EDR monitors process behavior, DLP protects documents, and MDM manages devices, they all focus on known applications and corporate-managed data flows. They fail to address the core characteristics of the Shadow AI use case:

• Business users are creating their own applications (scripts) on their workstations.
• They are manipulating company data they have legitimate access to.
• They are creating automations that affect other applications.

The intention is usually good: employees are innovating and generating value. However, they are simultaneously introducing significant security and operational risks.

Attempting to force business users into a traditional, secure software development pipeline (like GitHub) presents multiple hurdles:

• Technical Knowledge: It requires technical proficiency that the average business user lacks.
• Late-Stage Analysis: A secure pipeline only analyzes the code after it is submitted. The most significant risk occurs during development: the business user “playing” with the AI-generated code locally, where every execution could pose a risk.
• Scalability: Requiring a dedicated virtual environment for every employee is a monumental, non-scalable IT challenge.

Sentinel: Python Governance Without Blocking Innovation

The solution is not to block Python. That stifles innovation and forces the problem further into the shadows. The key is to provide visibility and granular control at the point of execution.

BotCity Sentinel shifts the technical aspects of governance to the endpoint, enabling organizations to manage the risk of Shadow AI without hindering business process improvement.

Sentinel provides continuous, deep-level monitoring across all Python execution, focusing on critical data points:

• Libraries: Import tracking.
• LLM Usage: Monitoring of LLM interactions.
• Communication: Inbound and Outbound network activity.
• Data Access: File Reading/Writing and Database Connections.
• System Interaction: Application Execution, Spreadsheet Processing, and Log Writing.
• Resource Use: Tracking of Computing Resources (CPU/RAM).

This comprehensive data gives teams full insight into every running Python script, including its location (machine), associated risks, and the ability to take immediate, precise action. Crucially, the entire solution operates 100% on-premises, guaranteeing data sovereignty and regulatory compliance.

Where Does Your Organization Stand?

We observe three distinct responses to the Shadow AI challenge:

1. Aware and Moving Fast: Companies in this category believe, “I have this problem, it’s massive, and I need visibility right now.” Their proposed action is to implement immediate endpoint monitoring and risk assessment.
2. Aware and Seizing the Problem: These companies state, “I know I have this problem, but I don’t know the scale or impact.” They plan to deploy a proof-of-concept to map the current Python estate and quantify the risk.
3. In Denial: Companies here claim, “Everything is blocked here. I’m sure we don’t have Shadow Python.” Their recommended action is to initiate a zero-footprint discovery scan to uncover what is actually running.

If you see your organization in one of these scenarios, the time to act is now. Get in touch to talk to us. We will show you exactly what is running on your endpoints and how to govern this new reality. Our specialists are ready to provide a deep, actionable dive into your current state. 

Don’t wait for the next incident. Let us empower you with the visibility and control needed to secure your digital environment and confidently govern your IT ecosystem.

Delve deeper into this discussion in the webinar

If this topic has raised concerns about your operation, it’s worth watching the webinar we’ve specifically designed for this subject.

In it, we show how risk is already present in endpoints and why BotCity Sentinel delivers the visibility and execution evidence that GRC and SecOps teams need to govern the use of Python more securely.