Secure compiler for AI-agent skills
Turn natural-language agent skills into structured, more deterministic Python programs, retaining generative model calls only where context or composition is needed. The compiled program can then be validated, hardened, and run with security safeguards.
online B2B Deep Tech / Infrastructure
From IBM Technology — AI skills security, Open AI Deployment Company & zero days at 01:50
Problem: Natural-language skills marketplaces are chaotic and can contain unsafe behavior, contradictions, prompt injection, instruction hacking, excessive tool permissions, and difficult-to-version prompts.
For: Organizations deploying AI agents and enterprises that need reusable skills to be safer, more reliable, verifiable, and easier to manage.
Products from this video
Amazon Bedrock Amazon Web Services (AWS) IBM Granite Mellea Mellea Skills Compiler Microsoft Azure Mythos Python Red Hat AI Inference Snowflake
Examples
- MELLEA: an IBM Research project intended to bring order and security to the agent-skills ecosystem.
Other takes on AI security infrastructure
- Agentic consent governance layer for autonomous AI systems
- Enterprise agentic last-mile identity and access control layer
- Identity-based security infrastructure for multi-agent AI systems
- AI-agent security and control infrastructure for enterprises
- A guardrail tool that enforces strict test-driven development and other engineering rules for AI coding agents
- AI-powered scam-interception and threat-intelligence service