Browse / Direct Prompt Injection
Direct Prompt Injection
Instructions typed into the model that override what the application told it to do. 53 records. Reference: OWASP LLM01 · ATLAS AML.T0051.
- Code and command execution 18
- LLM Prompt Crafting 12
- Authentication and access control 5
- Path traversal and file access 4
- SQL and query injection 3
- Cross-site scripting 3
- Other weaknesses 3
- Server-side request forgery 2
- Information exposure 1
- Unsafe deserialisation 1
- LLM Prompt Injection: Direct 1
- PraisonAI: Prompt-injection defense blocks only when 3+ detector families fire simultaneously; realistic single-vector injections pass through unblocked The prompt-injection scanner blocks only when several detector families fire together, so a single-pattern override instruction passes through unblocked.
- PraisonAI: AICoder Arbitrary File Write and Command Execution via LLM Tool Calls Prompt injection in the chat can make the coding assistant call unvalidated file-write and shell tools, writing arbitrary paths and running attacker commands.
- PraisonAI: Prompt Injection Defense Bypassed for HIGH-Level Threats
- PraisonAI: AgentMail webhook lacks signature verification, allowing unauthenticated message injection and sender spoofing
- Flowise Prompt Injection to RCE and SSRF via CSV/Airtable Agent Python Validator Bypass
- Xinference vulnerable to remote code execution via unsafe `eval()` in Llama3 tool-call parsing
- Flowise: CSV Agent Prompt Injection Remote Code Execution Vulnerability A prompt sent to a CSV agent chatflow makes the model emit Python that evades the blocklist and runs in the server's Pyodide-based sandbox.
- PraisonAI: Compute-bridged file tools allow shell command injection
- PraisonAI: Webhook signature verification skipped (fail-open) when secret unset, allowing forged inbound webhooks (WhatsApp & Linear bots)
- PraisonAI: Arbitrary File Read via `@file:` Mention Path Traversal Prompt-level @file mentions fall back to unchecked absolute paths, letting an attacker read any file the agent process can access.
- PraisonAI has an Arbitrary File Write in Python API
- RCE Vulnerability in Semantic Kernel Search Plugin
- Gemini CLI: Remote Code Execution via workspace trust and tool allowlisting bypasses
- Claude Code: Sandbox Escape via Symlink Following Allows Arbitrary File Write Outside Workspace
- Flowise: Airtable_Agent Code Injection Remote Code Execution Vulnerability
- Flowise: APIChain Prompt Injection SSRF in GET/POST API Chains Unauthenticated users can inject API documentation through the question, steering the chain's LLM-built requests to arbitrary internal or external URLs.
- Flowise: Remote code execution vulnerability in AirtableAgent.ts caused by lack of input verification when using `Pandas`. User questions are inserted unsanitised into the prompt, so a crafted prompt makes the model generate Python that the Pyodide runtime executes.
- PraisonAI: Unauthenticated Information Disclosure of Agent Instructions via /api/agents in AgentOS
- PraisonAI: Hardcoded `approval_mode="auto"` in Chainlit UI Overrides Administrator Configuration, Enabling Unapproved Shell Command Execution
- PraisonAI: Coarse-Grained Tool Approval Cache Bypasses Per-Invocation Consent for Shell Commands
- PraisonAIAgents: Arbitrary File Read via read_skill_file Missing Workspace Boundary and Approval Gate
- PraisonAIAgents has an OS Command Injection via shell=True in Memory Hooks Executor (memory/hooks.py)
- PraisonAIAgents: Environment Variable Secret Exfiltration via os.path.expandvars() Bypassing shell=False in Shell Tool
- PraisonAI Vulnerable to Arbitrary File Write / Path Traversal in Action Orchestrator
- PraisonAI Has Sandbox Escape via shell=True and Bypassable Blocklist in SubprocessSandbox
- Langflow has Remote Code Execution in CSV Agent A prompt-injected message makes a CSV agent call a Python REPL tool because the dangerous-code flag is hardcoded on, so the model's chosen actions run on the host.
- OpenClaw Command & Control via Prompt Injection
- LangChain serialization injection vulnerability enables secret extraction
- Open WebUI vulnerable to Stored DOM XSS via prompts when 'Insert Prompt as Rich Text' is enabled resulting in ATO/RCE
- Lobe Chat Desktop vulnerable to Remote Code Execution via XSS in Chat Messages
- Cross-Site Scripting via Prompt Manipulation in Lenovo AI Chatbot
- Code to Deploy Destructive AI Agent Discovered in Amazon Q VS Code Extension
- Malware Prototype with Embedded Prompt Injection
- Living Off AI: Prompt Injection via Jira Service Management
- Data Exfiltration via Agent Tools in Copilot Studio
- EchoLeak: Zero-Click Prompt Injection Targeting M365 Copilot for Data Exfiltration
- LlamaIndex vulnerable to Creation of Temporary File in Directory with Insecure Permissions
- Rules File Backdoor: Supply Chain Attack on AI Coding Assistants
- PandasAI interactive prompt function Remote Code Execution (RCE)
- Flowise OverrideConfig security vulnerability
- Langchain SQL Injection vulnerability Prompt-injected text makes the LLM generate Cypher queries that GraphCypherQAChain runs unchecked, letting attackers read, change or delete graph data.
- vanna vulnerable to remote code execution caused by prompt injection
- Vanna prompt injection code execution User text passed to the ask method can override the prompt that generates visualisation code, so the model emits Python that the library then executes.
- llama-index-core Prompt Injection vulnerability leading to Arbitrary Code Execution
- Hacking ChatGPT's Memories with Prompt Injection
- China-Based Artificial Intelligence Companies Conducting Industrial-Scale Distillation Campaigns Against U.S. AI Companies
- Langchain Server-Side Request Forgery vulnerability Prompt injection makes the service fetch an attacker-chosen URL, and the retrieved content can then be injected into downstream tasks.
- Langchain SQL Injection vulnerability Prompt text steers the chain to build SQL that the SQL service runs unsafely, letting an attacker execute arbitrary commands against the database.
- pandasai vulnerable to prompt injection
- PandasAI vulnerable to arbitrary code execution Natural-language prompts can evade the jailbreak keyword check, so model-generated code runs on the host.