Frontend developers know one golden rule: never render untrusted user input as HTML without properly sanitizing it. Every framework has its own APIs and escape hatches where this rule matters.

Yet, as we're integrating Large Language Models (LLMs) into our applications, we often make a potentially dangerous mistake: we treat AI-generated output as a trusted source.

That should be fine, right? Well, not automatically…

Software changes. Dependencies evolve. New requirements appear. A Symfony application that can adapt without major disruption gains a powerful long-term advantage — this talk explores architectural choices that keep core business logic stable while technology evolves.