AI Red Team Engineering notes, research references, and field notes on AI systems, responsible AI, governance, and compliance.
Global search

Search AI Red Team

Search across the complete knowledge base.

Enter at least two characters to begin.
↑↓ select Enter open Browse the full research archive
NVIDIA AI Red Team · December 16, 2024

Sandboxing Agentic AI Workflows with WebAssembly

guide Agent SecurityAI EngineeringPrompt Injection
Why it matters

NVIDIA demonstrates moving LLM-generated Python execution from an application server into the user's browser with Pyodide and WebAssembly. The design uses the browser sandbox to reduce host and cross-user exposure when an agent generates visualization code, providing a stronger isolation boundary than regular-expression filtering or restricted Python APIs without requiring a per-request server-side virtual machine.

My takeaway: Treat browser-based WebAssembly as one containment layer, not permission to execute arbitrary model output. Keep secrets and privileged APIs out of the runtime, restrict network and DOM capabilities, cap CPU and memory, validate exported files and results, and test prompt-injected code against the actual browser policy. Use stronger isolation when generated code must access sensitive data or tools.
Keep exploring

More curated notes connected through Agent Security and AI Engineering.

OWASP GenAI Security Project · guide

OWASP Top 10 for Agentic Applications for 2026

OWASP's community guide organizes agentic-system risk into ten categories, including goal hijacking, tool misuse, identity and privilege abuse, memory poisoning, insecure inter-agent communication, cascading failures, and rogue-agent behavior. It provides a shared taxonomy and mitigation starting point rather than a certification checklist or evidence that a deployed system is secure.

Microsoft Security Blog · guide

AI vulnerability research: measure reproducible findings and completed fixes

Microsoft’s FORGE account describes the work between a model’s vulnerability claim and a useful repair: reusable builds, duplicate removal, reachability checks, project-specific verification, reproducible triggers and regression tests. Structured rejection reasons help improve later searches. The useful operational measure is the flow of findings that survive verification and reach a fix, rather than the number of candidates generated. Reported successful-case costs exclude parts of screening, failed attempts and human work, so they are not the total cost of operating this pipeline.

OpenAI News · framework

Frontier training safety cases: connect evidence to enforced pause and rollback controls

OpenAI proposes training-run safety cases combining alignment evaluations, containment and monitoring with explicit operational ownership. Concrete measures include immutable transcripts, held-out incident tests, checks for evaluation gaming, response deadlines and fail-closed monitoring. Independent internal challenge, leadership vetoes and tracking downstream uses support stopping a run and reversing affected work. The article describes recommendations still being implemented, rather than audited proof that every safeguard already operates or that residual risk has been eliminated.