WebAssembly, Python, and the Future of AI Sandboxing

AI workloads increasingly require secure, fast, and portable environments for running code, without the overhead of containers or VMs. WebAssembly (Wasm) offers a lightweight sandbox that runs anywhere: on servers, on the edge, and even directly in the browser.

In this talk, we’ll explore how Python and WebAssembly work together to safely execute untrusted or distributed workloads, what makes Wasm’s capability-based security model unique, and why it’s becoming a key building block for modern AI systems. We’ll also look at how Wasmer expands what’s possible for Python developers beyond the web.

Transcript

1. WebAssembly, Python, and the Future of AI Sandboxing Syrus Akbary Feb 2026 Rethinking safe code execution for the AI era
2. Who am I?
6. I want to make it universal! Circa 2018
8. Wait. What the f*** is WebAssembly? Well… maybe I should have started here
10. such cool WOW very Wasm
11. What is Wasmer? Wasmer is a WebAssembly Platform • Runtime (Analogy: Docker) • Registry (Analogy: Docker Hub or NPM) • Edge (Analogy: AWS Lambda) Wasmer runs applications as Wasm modules instead of containers or virtual machines
12. • Do not share kernel or OS • Poor performance • Very low density • Shares a kernel, but not the OS or binary • Still very heavy- weight • Low density • Shares a kernel, and the binary executables • Fully sandboxed • High density
13. Modern systems increasingly execute code we didn't write The question is no longer if — but how to execute safely and ef f iciently
14. Why this matters We need a new Execution Primitive More Ephemeral Short-lived workloads More Multi-tenant Shared infrastructure More Untrusted AI-generated code
15. Is this not a Python talk?
16. How Python is using Wasm?
17. Browser- f irst Pyodide is the most popular framework. Is Python compiled to Wasm via Emscripten. Used by: • JupyterLite • PyScript(.com) • Anaconda Cloud • Cloud f lare Optimized for JS execution
18. Server- f irst You can run pyodide in Node.js (very heavy). Or you can run it with Wasmer. Used by: • Wasmer Edge (hosting) • AI Companies Optimized for density and sandboxing
19. But Pyodide has constraints No Raw Sockets Limited by the browser's networking model, preventing direct socket access. Limited POSIX Compatibility constraints mean certain system-level operations are restricted. No Multithreading Python code executes on a single browser thread, limiting parallel computation. No Subprocesses/Greenlets Prevents parallel processing within the Python environment for tasks like I/O concurrency. Not designed for server workloads (even though Cloud f lare uses it)
20. Why the Python+Wasm on the server matters …specially for AI workloads Sandboxing is now a core infrastructure requirement • Untrusted Python code execution • Fast response/startup times • High Density
21. The emerging Sandboxing Pattern Resource Limits • CPU, memory, execution time constraints Capability-based access • Filesystem permissions • Networking permissions High performance • High density • Fast startup times
22. Wasmer’s approach to Python in Wasm
23. Browser- f irst Optimized for JS execution Optimized for density and sandboxing Server- f irst ❤ Enables • 👨💻 Zero modi f ications • 🚀 Run in any environment and OS • ⚡ Extremely lightweight Without limitations • ✅ Greenlets • ✅ Multithreaded • ✅ Subprocess
24. Wasmer Clouds vs Docker Clouds Containers • Heavy isolation overhead • Slow startup times • Low tenant density WebAssembly • Shared sandboxed execution • Extremely fast startup times • High tenant density
25. Demo time! Using Python with Wasmer + Clouds
26. One year free of Wasmer Pro account 1. Create an account https://wasmer.io/ 2.Reach syrus@wasmer.io with your username
27. Thank you 🙏 @syrusakbary / @wasmerio