Function Parameter Injection

Function Parameter Injection is an attack technique where an adversary manipulates the arguments passed to AI function calls, causing unintended operations, unauthorized data access, or execution of malicious actions through otherwise legitimate enterprise integrations.

Modern AI assistants frequently interact with external applications through structured function calling. Rather than exploiting the function itself, attackers manipulate the parameters supplied by the AI model to influence how trusted business applications behave.

Successful Function Parameter Injection can result in unauthorized transactions, excessive data retrieval, privilege misuse, or modification of enterprise systems despite the underlying function operating exactly as designed.

Function calling enables AI models to execute structured operations such as querying databases, updating records, scheduling meetings, processing payments, or interacting with internal APIs.

Attackers craft prompts that influence the AI’s reasoning, causing it to generate unexpected or malicious function parameters. Because many enterprise systems trust AI-generated arguments without sufficient validation, manipulated parameters can trigger unintended business actions.

Unlike prompt injection, the attack focuses on corrupting structured inputs after the AI has already selected a legitimate function

1. An attacker submits crafted instructions to an AI assistant.
2. The AI selects an appropriate enterprise function.
3. The attacker influences the generated function parameters.
4. The external application executes the request.
5. Unauthorized actions occur using valid enterprise integrations.

• AI Function Calling APIs
• Enterprise AI Assistants
• AI Agents
• CRM Platforms
• ERP Systems
• Internal APIs
• Business Automation Platforms

• Unauthorized business transactions
• Data overexposure
• Privilege misuse
• Manipulation of enterprise workflows
• Regulatory compliance violations
• Financial loss
• Loss of confidence in AI automation

Redbot Security evaluates Function Parameter Injection through controlled testing of AI-enabled enterprise integrations.

Testing includes:

• Parameter manipulation
• Authorization validation
• Input boundary testing
• Business logic testing
• Structured payload analysis
• Function execution monitoring

Monitor for:

• Unexpected parameter values
• Excessive function arguments
• Unauthorized resource identifiers
• AI-generated requests outside expected patterns
• Abnormal API activity
• Unexpected business transactions initiated by AI

AML.T0051 – Prompt Injection AML.T0011 – AI System Abuse

LM01 – Prompt Injection LLM06 – Excessive Agency