BrightOnLABS @ Pytorch Conference

Agentic Systems Are Scaling, Their Security Isn’t (Yet)

At the PyTorch Conference 2025 (hosted in San-Francisco) , one trend was impossible to miss:
AI agents and agentic workflows are rapidly moving into production.

But that shift brings a new reality.
Traditional, deterministic security models don’t apply to systems where actions emerge from non-deterministic reasoning, cross-agent interactions, and tool-enabled execution paths.

Introducing BrightOnAGENTIC

During the showcase, we presented BrightOnAGENTIC, our agentless security engine designed for this new era.

• It:
  • Auto-discovers agent frameworks
  • Maps identities, permissions, and toolchains as a graph
  • Models blast radius from agent decisions
  • Prioritizes risks unique to agentic behavior
  • Generates actionable remediation commands
  • Live-monitors agentic systems at runtime

The message resonated, Agentic Security is quickly becoming its own category.

The Road Ahead

As agentic systems continue to evolve, security must evolve with them — from static, rule-based protection to dynamic, context-aware, graph-based reasoning.
That’s the mission behind BrightOnLABS and our platform BrightOnX.ai: securing the future of AI-driven systems with real-time visibility, explainability, and trust.

Finally, we want to thank the Linux Foundation and PyTorch team for featuring us and giving visibility to early builders in this new field.

If you’re building or deploying agentic systems, we’d love to connect:
contact@brightonx.ai
contact@brightonlabs.ai 

Cheers
BrightOnLABS