Models openai for Prompt injection

[mbaluda]

No description provided.

[Copilot]

Pull request overview

This PR adds support for detecting prompt injection vulnerabilities (CWE-1427) in Python code that uses the OpenAI package. It introduces a new security query that identifies when user-provided input flows into system prompts or developer messages, which could allow attackers to manipulate AI model behavior.

Key changes:

• Creates a new prompt injection detection query with taint-tracking configuration
• Adds OpenAI and Agent framework models to identify vulnerable sink points in the API
• Includes comprehensive test cases covering various OpenAI API patterns

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 12 comments.

Show a summary per file

File	Description
python/ql/test/query-tests/Security/CWE-1427-PromptInjection/openai_test.py	Test cases for OpenAI API patterns including chat completions, responses, and realtime connections
python/ql/test/query-tests/Security/CWE-1427-PromptInjection/agent_instructions.py	Test case for Agent SDK instructions parameter
python/ql/test/query-tests/Security/CWE-1427-PromptInjection/PromptInjection.qlref	Query reference file pointing to the security query
python/ql/test/query-tests/Security/CWE-1427-PromptInjection/PromptInjection.expected	Expected test results showing edges, nodes, and alerts
python/ql/src/Security/CWE-1427/PromptInjection.ql	Main query implementation for prompt injection detection
python/ql/src/Security/CWE-1427/PromptInjection.qhelp	Documentation describing the vulnerability and recommendations
python/ql/lib/semmle/python/security/dataflow/PromptInjectionQuery.qll	Taint-tracking configuration for the query
python/ql/lib/semmle/python/security/dataflow/PromptInjectionCustomizations.qll	Defines sources, sinks, and sanitizers for prompt injection
python/ql/lib/semmle/python/frameworks/OpenAI.qll	Framework models for OpenAI and Agent APIs identifying vulnerable parameters
python/ql/lib/semmle/python/Frameworks.qll	Registers the OpenAI framework models

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

There is a grammatical issue in this sentence. "and or" should be either "and/or" or simply "or". The recommended wording would be: "User input used in developer message or system prompt can allow for Prompt Injection attacks."

Suggested change

	* @name User input used in developer message and or system prompt
	* @description User input used in developer message and or system prompt can allow for Prompt Injection attacks.
	* @name User input used in developer message or system prompt
	* @description User input used in developer message or system prompt can allow for Prompt Injection attacks.

[Copilot]

The example reference file path is incomplete. "examples/TODO.py" should be replaced with an actual example file path or removed if the example doesn't exist yet.

Suggested change

<sample src="examples/TODO.py" />