Salus

Roadmap & Position in AI Safety

Real-time guardrails that validate AI agent actions before execution to prevent unsafe outputs.

Company Overview

Builds real-time, ML-driven guardrails that validate AI agent actions before execution, using hybrid LLM-based evaluators and evidence grounding to prevent unsafe or hallucinated outputs.

What They're Building

The company's public product roadmap & what they're committed to building.

Product website describes core capabilities: real-time action validation via API wrapper, evidence cache for grounding agent decisions, guided retries with structured feedback for blocked actions, and policy configuration in YAML/Markdown/plain English. Benchmarked at 52% reduction in misalignment on ODCV-Bench across 12 frontier models, with 58% recovery rate on blocked actions. Integration via pip install and a few lines of code.

Latest Intelligence

Zeitgeist tracks private signals to determine where the company is heading strategically.

No Signals Yet

Competitors

Guardrails & Validation

Guardrails AI (open-source output validation), NVIDIA NeMo Guardrails (dialogue safety rails).

LLM Security

Lakera (LLM firewall/prompt injection defense), Strong Intelligence (AI validation platform).

LLM Evaluation

Patronus AI (LLM evaluation & testing), Calypso AI (model security & governance).

Open Source

LangChain safety modules, various agent safety frameworks.

Salus

's Moat:

Pre-execution validation (vs. post-execution monitoring) is architecturally upstream of competitors like Guardrails AI. 52% misalignment reduction across 12 frontier models with 58% action recovery rate is measurable, testable performance. pip install with YAML config lowers adoption friction below enterprise-sales-dependent competitors.

How They're Leveraging AI

AI Use Overview:

Using hybrid LLM action validation with evidence grounding, evidence-based observability for audit trails, and autonomous agent self-repair via guided retries.

More Similar Companies

Arena (formerly LLMArena)

Crowdsourced human-preference benchmarking platform for LLMs and generative AI models.

Neutral third-party evaluation becomes critical infrastructure as model proliferation outpaces any single lab's ability to grade itself credibly.

Ashr

Catches AI agent failures before users see them by stress-testing across text, voice, and images.

AI agents are shipping to production faster than anyone can test them. Ashr generates synthetic users that stress-test agents across text, voice, and images before real users hit the failure modes.

Cajal

Deploys AI mathematicians that formally verify proofs, grounding outputs in truth not guesses.

LLMs hallucinate. Lean proves things. Cajal pairs LLMs with formal verification so every mathematical result is machine-checked, starting with quantum computing and finance where a wrong proof costs real money.

Cascade

Evaluates and certifies AI agents for safe deployment with red teaming and formal guarantees.

Red teaming and guardrails exist as separate tools. Cascade combines them into one platform with adaptive scaffolding that learns from production runs, already deployed across legal reasoning and customer support agents. The CEO researched graph reasoning and agentic safety at UC Berkeley's BAIR Lab.