AI security infrastructure

Build security, identity, permission, monitoring, and control systems for autonomous AI agents.

2 creators independently cover this, across 3 variations.

online B2B

Sources: 20VC with Harry Stebbings · IBM Technology

AI security controls for autonomous agents

20VC with Harry Stebbings — Leo Aschenbrenner's Situational Awareness Blows Up | Moonshot AI Raises $3.5B at $35B

Build identity, permission, monitoring, interception, and kill-switch infrastructure for agents with the ability to make decisions.

Problem: Agents can access systems and take actions without sufficient security guardrails.

For: Enterprises deploying AI agents.

Examples

  • Agent identity
  • Privileged identity treatment
  • Inline interception
  • Kill switches

AgentOps tooling for enterprise AI deployments

IBM Technology — Agent control planes & OpenAI model solves Erdős

Tools that adapt software development and operations workflows to probabilistic agents.

Problem: Lack of governance, safety, trust, observability, identity, compliance, cost control, and reliable evaluation.

For: Enterprises operating many AI agents across business units.

Examples

  • Agentic control planes
  • Evaluation harnesses
  • Agent observability and optimization tools

Defensive self-spreading vulnerability analysis

IBM Technology — Can you social engineer an AI? Plus: AI worms and the nonhuman identity problem

Use the methodology of an AI worm to distribute an AI agent that identifies vulnerabilities and reports how to fix them rather than exploiting them.

Problem: Finding device-specific vulnerabilities across an environment.

For: Organizations seeking to identify and remediate vulnerabilities.

Examples

  • An agent that reasons through vulnerabilities on each device and recommends patches