Map / Runtime & Guardrails / AariaSec
AariaSec
aariasec.com ↗ · Open on the map
Runtime & GuardrailsObservability & Governance
Capabilities
Behavioral security for AI agents already running in production: detects hijacked or drifted agents that still hold every credential they were granted, so no policy is violated and nothing else fires. Learns a per-agent baseline — tool usage, data destinations, timing, token patterns — and flags when an agent stops behaving like itself. Never stores prompt or response text.
Sensors / integration
Content-free telemetry only: SHA-256 fingerprints of prompts and responses captured at the boundary, plus tool names, latencies, call bursts, and egress destinations. Per-agent learned baselines rather than generic thresholds; fleet-level drift detection; a debate panel of independent models argues each alert before it surfaces.
Protections
Blocks destructive tool calls (deletion, data wiping) by default, with optional automatic containment. Enterprise tier adds kernel-enforced sandboxing via macOS Seatbelt, Linux Landlock plus seccomp, and Windows AppContainer.
Differentiator
Attacks the gap credential-based controls miss — a compromised agent keeps legitimate access, so identity and policy tooling stays silent. Because the telemetry holds no content, AariaSec can publish PRAMANA, an open agent-compromise detection benchmark, including the attack classes where it scores worst: nothing in it can leak. Pre-revenue with three signed SMB pilots; San Jose based.
Team & funding
- Founders
- Ajith Chandran, Liji Chalatil
- Investors
- Self-funded (raising)
Other Runtime & Guardrails companies
ActaClad · ActiveFence · Acuvity · AGAT Software · AgentFox · AI Aramid · AI Prompt Shield · Aiceberg · Aim Security · AIR · Aira Security · Airrived · See all →
Last updated · Get the “Listed on” badge