𝗔𝗦𝗜𝟬𝟱: 𝗨𝗻𝗲𝘅𝗽𝗲𝗰𝘁𝗲𝗱 𝗖𝗼𝗱𝗲 𝗘𝘅𝗲𝗰𝘂𝘁𝗶𝗼𝗻 (𝗥𝗖𝗘) emerges when untrusted promp…

𝗔𝗦𝗜𝟬𝟱: 𝗨𝗻𝗲𝘅𝗽𝗲𝗰𝘁𝗲𝗱 𝗖𝗼𝗱𝗲 𝗘𝘅𝗲𝗰𝘂𝘁𝗶𝗼𝗻 (𝗥𝗖𝗘) emerges when untrusted prompts, files, tool responses or generated content cross the boundary into a shell, interpreter, build pipeline or runtime. Once that boundary disappears, a prompt-injection problem can become system compromise.

Defensive priorities:

• Treat all model output as untrusted data
• Never pass free-form output directly to a shell or interpreter
• Use strict schemas and allowlisted operations
• Execute in isolated, ephemeral sandboxes
• Deny unnecessary network egress
• Require approval for high-impact execution

Where does generated content cross into execution in your architecture?

#AgenticAI #RCE #ApplicationSecurity #Sandboxing #AISecurity #OWASP #SecureByDesign unexpected code execution, RCE, sandboxing, code interpreter, command injection, unsafe deserialization, model output validation, egress control, OWASP ASI05