WillItSurvive
YC Winter 2026 · United States

Salus

Runtime guardrails that check an AI agent's actions before it executes them.

Protection under test
Assessed 27 July 2026 · Evidence as of 27 July 2026
Something worth protecting exists and is being tested right now.
On the recordSince 27 July 2026 · Checkable from Q1-Q2 2027 · Every call on the record Checkpoint: Target window to land 2-3 named regulated-enterprise customers as system-of-record for agent policy
The call

Salus sells a guardrail layer that big clouds and OpenAI are now building themselves, with no named customer yet proving anyone needs the extra layer.

Salus builds guardrails, which are safety checks that run between a user and an AI system to catch bad outputs or stop harmful requests. The problem is that Microsoft, Amazon, and OpenAI have all started shipping their own guardrails directly into their AI platforms in 2025 and 2026, and they have already bought up Salus's competitors like Lakera and Promptfoo. Salus has no famous customers that the public knows about and no special advantage that the big platforms cannot copy. The one thing that could save Salus is if it either proves it works better across many different AI systems at once, or if it focuses on one specific industry like banks or hospitals where safety rules are stricter and harder to get right than what the big platforms offer.

Why
01
AI pressure: OpenAI Guardrails and its Promptfoo acquisition. OpenAI now ships native policy checks and bought a rival's guardrail tool, folding Salus's whole pitch into the model layer. Could AI do the core job? Yes, likely.
02
Bigger players: Microsoft Azure AI Foundry and AWS Bedrock Automated Reasoning. Both clouds now ship built-in agent guardrails, so buyers can get policy checks free where they already run their agents. Could a giant copy it? Yes, easily.
03
What protects it. A one-line, platform-agnostic integration lets it sit above Azure, AWS, and OpenAI stacks instead of being locked to one.
What would change this call
Salus announces a named regulated-industry customer (finance or healthcare) by Q2 2027. Would show it can win buyers against free platform-native guardrails.
Salus is acquired or quietly folds its GTM hiring within 6-12 months. Would confirm the acquihire/consolidation trajectory judges warned about.
Salus publishes an independently verified self-repair or cross-platform benchmark, not vendor-authored. Would show its 58% self-repair claim is a real lock-in, not marketing.
Key numbers
  • 58% of blocked agent actions self-repaired, per vendor benchmark (2026-02-06)
  • Claims 60% lower cost on tau2-bench, vendor benchmark (2026-02-06)
  • $4M raised per company blog post (2026-05-29)
Named risks
  • Core product function is being replicated free by Microsoft, AWS, and OpenAI natively. Severity: high
  • No disclosed customers, revenue, or named enterprise logos as of mid-2026. Severity: high
  • Category consolidating fast via four acquisitions within about nine months, closing independent-vendor window. Severity: high
  • Conflicting funding figures ($500K to $4M) suggest unclear or unstable capital position. Severity: medium
  • Broader guardrails ecosystem faces supply-chain attacks, as seen in May 2026 Mini Shai-Hulud trojanized package. Severity: medium
Rivals, as read in July 2026
RivalDirectionEvidence
Microsoft Azure AI Foundrylosing toShipped native agent guardrails in preview by 2026-05-14, replicating Salus's core function
AWS Bedrocklosing toAdded Automated Reasoning checks for formal-rule output validation in 2026
OpenAIlosing toLaunched OpenAI Guardrails and acquired Promptfoo in March 2026, vertically integrating the category
WitnessAIlosing toReportedly raised $58M as of 2026-04-30, far outpacing Salus's disclosed early-stage funding
Portkey, Lakera, CalypsoAI, Promptfoo (acquired specialists)contestingAll four acquired by larger security/cloud vendors between Sept 2025 and April 2026, consolidating the niche
Questions people ask
Is Salus profitable in 2026?
No. As of mid-2026 Salus has disclosed no revenue, customer count, or named enterprise logos. Its only public traction metrics are self-reported benchmark results (58% self-repair rate, 60% cost reduction claims from its Feb 6, 2026 YC launch), not commercial revenue.
Could AI platforms replace Salus?
Largely yes. Microsoft Azure AI Foundry shipped native agent guardrails in preview by May 14, 2026, AWS Bedrock added Automated Reasoning checks, and OpenAI launched its own Guardrails layer plus acquired Promptfoo in March 2026, replicating Salus's core policy-check function inside the platforms it depends on.
Who founded Salus and when did it launch?
Salus was founded by Stanford CS roommates Kevin Pan and Vedant Singh as a Y Combinator Winter 2026 batch company, and publicly launched via YC Launch on February 6, 2026, offering runtime guardrails that validate AI agent actions before execution.
How much funding has Salus raised?
Sources conflict: one aggregator cites a $500K YC-led pre-seed, another cites $3.7M, and a May 29, 2026 blog post claims $4M raised. No primary press release or SEC filing reconciles these figures, so the exact amount remains unconfirmed as of mid-2026.
Who is Salus's biggest competitor?
Salus competes against both better-funded specialists like WitnessAI (reportedly $58M raised, as of April 30, 2026) and platform giants Microsoft, AWS, and OpenAI, all of which shipped native agent guardrail features in 2025-2026, directly threatening Salus's proxy-layer positioning.
Who owns Salus?
Salus is a privately held, venture-backed company with Y Combinator as a batch investor and backer (partner Ankit Gupta cited), co-founded and presumably majority-owned by Kevin Pan and Vedant Singh as of the company's 2026 YC W26 launch.
Timeline
  • 2026-06-01Salus posts Founding GTM/BDR job listing in San Francisco, signaling shift to commercialization
  • 2026-05-29Blog post claims Salus team raised $4M, conflicting with earlier funding figures
  • 2026-05-14Microsoft Azure AI Foundry ships native agent guardrails in preview
  • 2026-04-30Portkey acquired by Palo Alto Networks, continuing guardrails-market consolidation
  • 2026-03-01OpenAI acquires Promptfoo and ships first-party OpenAI Guardrails layer
  • 2026-02-06Salus publicly launches via YC Launch, claiming 58% self-repair and 60% cost reduction benchmarks
  • 2025-09-01Lakera acquired by Check Point and CalypsoAI acquired by F5, early consolidation signal
Revision history
27 July 2026
First assessed. Protection under test. The text above is fixed as of this date and is not edited afterwards.
In the same band
Sources