Secure your data computations efficiently
summary
Claims a new restricted language, Starlarky, that isolates computations at the language level and prevents common abuse by design: no filesystem or network access, no while or recursion, deterministic execution with bounded objects and immutable iterators. Demonstrates embedding in Java and Python using the Larky runtime, a regex/redaction demo with VGS Vault, conversion from Python via pi2star, and a working gRPC function-as-a-service server built with Maven (Docker optional). Demos run successfully and Starlarky is in production at VGS.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:10Data security importance
- 4:20Attack surface in computations
- 5:43State-of-the-art solutions
- 9:15Introducing Starlarky
- 13:31Demo embedding Starlarky
- 29:56Q&A session
- 34:51Workshop overview
- 53:43Building function service
- 1:07:25Contributing to Starlarky
- 1:12:04Closing and resources
related talks
32mBazel Rules
46mGuarding the LLM Galaxy: Security, Privacy, and Guardrails in the...
44mSecuring LLM-Powered Apps: Overcoming Security and Privacy Challenges
01h 00mAI Safety through Programming?
50mUsing language-theoretics and runtime visibility to align AppSec with DevOps
50mSecuring LLM-Powered Applications: Overcoming Security and Privacy Challenges
51mSecuring LLM-Powered Applications: Overcoming Security and Privacy Challenges
38mBuilding Secure Systems in Haskell