This document provides a clear overview of the key security considerations when working with AI-assisted development in BoldCreate. It explains the risks, built‑in protections, and best practices you should follow to stay safe while benefiting from AI‑powered workflows.
Why Security Matters
AI agents in development tools can automate tasks, modify files, and run commands. While this greatly improves productivity, it also increases the risk of accidental or malicious actions. Understanding how these systems work—and the protections around them—helps you use them safely.
Execution & Access Risks
When you use AI agents in BoldCreate, they can perform:
File Operations
- Create, edit, or delete files in your workspace.
- Trigger tools or scripts (like build or watch tasks) when files change.
Terminal Commands
- Run shell commands with your user privileges.
- Install software or modify system configurations.
Extensions & MCP Servers
- MCP (Model Context Protocol) servers can access all local files.
- They can run arbitrary code and interact with system resources.
Because these actions carry risk, you must be aware of what’s being executed on your behalf.
Supply Chain Risks
Modern AI development relies on many external components. Each one introduces trust concerns.
MCP Server Integrity
- Third‑party MCP servers might contain vulnerabilities or harmful code.
- They may not have formal security reviews.
External Tool Dependencies
- AI agents may call external command‑line tools.
- These tools could be outdated or compromised.
Update Channels
- MCP server updates may come from untrusted sources.
- Malicious updates could introduce harmful behavior.
Auto‑Approval Risks
Auto‑approval speeds up development, but reduces visibility.
Edit Auto‑Approval
- Automatically applies file edits.
- Risk: Sensitive files could be modified without review.
Terminal Auto‑Approval
- Allows commands to run without confirmation.
- Risk: Destructive commands could execute silently.
Overall Tool Auto‑Approval
- Enables all operations without asking you.
- Risk: Arbitrary code execution, unintended workspace changes.
Use auto‑approval features carefully—only when you understand the impact.
Information Exposure Risks
AI systems handle a lot of context, which may include sensitive information.
Context Sharing
- Files, environment variables, and configuration details may be sent to the model.
- API keys or proprietary code can be accidentally exposed.
Tool Interactions
- Sensitive information retrieved from one tool can be inadvertently shared with another tool.
External Content
- Untrusted content from external sources can be introduced into your workspace through tool operations and file edits, potentially leading to data leakage.
Prompt Injection Risks
Malicious instructions hidden inside comments, files, or tool outputs can trick the AI.
Examples of Attacks
- Hidden text telling the AI to delete files.
- Commands embedded in user‑generated content (e.g., GitHub READMEs).
Possible Impacts
- Data exfiltration: Sending your data to unknown servers.
- Context contamination: Misleading the AI.
- Tool chaining attacks: Malicious output flows through multiple tools.
- External data exploitation: Using untrusted web content to alter behavior.
Built‑In Security Protections
- BoldCreate includes multiple safety layers to reduce the risk of harmful actions.
- BoldCreate does not use your workspace data, code, or context for model training.
Trust Boundaries
- Workspace Trust: Restricts execution until you trust the folder.
- Extension Publisher Trust: Installs the extensions which are trusted by the user.
Controlled Scope
- AI tools can access only your current workspace.
- Tool picker lets you enable/disable tools.
- Temporary session permissions prevent long‑term risks.
Permission Management
- Terminal commands require explicit approval.
- Tools require approval at session, workspace, or user level.
Transparency
- You review all AI‑proposed file changes before applying them.
- Clear messages show when auto‑approvals occur.
- Warning banners appear for advanced or risky modes.
Secrets Management
- Sensitive values are stored securely.
- OAuth-based authentication protects MCP connections.
Note: Read our transparent legal policies and agreements in the Syncfusion Legal Center to ensure full compliance with and understanding of Syncfusion’s terms and conditions.
Developer Responsibilities
Even with strong protections, you play a key role in staying secure.
Review all edits
- Always check changes to config files, scripts, build tools, and security‑related files.
- Rely on source control to track and revert changes.
Review approvals
- Don’t approve commands or operations you don’t understand.
- Regularly review your auto‑approval settings.
Verify MCP servers
- Install MCP servers only from trusted sources.
- Enable them only when needed.
Be careful with untrusted codebases
- Open unknown projects in restricted mode.
- This disables agents and prevents prompt injection attacks.
Summary
AI-powered development is powerful but introduces new security challenges. BoldCreate helps protect you with trust boundaries, controlled scope, permissions, transparency, and secret management. As a developer, staying alert, reviewing actions, and only enabling what you trust will help you maintain a safe and productive workflow.