Something quietly changed in how security tooling gets evaluated. The first question about a product is no longer typed into a search box — it is asked of an assistant. “Can this tool produce evidence my auditor can verify independently?” And the assistant answers from whatever documentation, skill package or tool schema it can reach.
That creates a failure mode with no equivalent in the pre-assistant era: a product can be perfectly capable and still be described inaccurately by the surface an evaluator happens to query first. The report says one thing, the packaged skill says another, and the buyer never learns which one is current.
Nsasoft US LLC’s release of NSAuditor AI Enterprise 0.33.1, Community Edition 0.2.38 and agent-skill 0.2.36 is aimed directly at that gap.
The capability, stated identically everywhere
NSAuditor AI Enterprise supports RFC 3161 trusted timestamping as an opt-in capability. Set the NSAUDITOR_TSA_URL environment variable to a Time-Stamp Authority you choose, and every compliance artifact a scan produces — the framework report, the scope attestation, the chain of custody — carries an RFC 3161 .tsr sidecar. An assessor verifies it with stock openssl, offline, against the authority’s own certificate chain, with no vendor software in the path.
As of this release, the seven framework reports, the agent-skill package an assistant loads, and the Community Edition README all describe that capability the same way — including the parts a careful buyer needs: it is opt-in, it has no default ever, it requires openssl on the scanning host, and leaving the variable unset means nothing is timestamped and nothing is sent.
Why an AI-facing surface deserves the same care as the product page
A skill package is not marketing collateral; it is an interface. When an assistant loads it, its sentences become the assistant’s answers, quoted with the assistant’s confidence and no visible provenance. A stale sentence in a skill file is therefore not a documentation defect — it is a wrong answer delivered to a buyer who has no way to tell it is wrong.
Agent-skill 0.2.36 also adds an explicit status section for suppression signing, so an assistant asked about it returns the same answer the shipped product would: the signer backends and record schema exist, but no shipped entry point signs a suppression record today. Stating a boundary plainly is what makes the surrounding claims worth believing.
What the MCP server exposes
NSAuditor AI ships an MCP server, so an assistant can drive the product rather than only describe it: host and cloud scanning across AWS, Azure and GCP, findings and vulnerability retrieval, service probing, plugin enumeration, and compliance_matrix — which returns the shipped coverage matrix for any of the seven frameworks, derived from the framework data at call time and failing closed rather than answering from memory when it cannot read that data.
That design choice is the same instinct as the release itself: remove the opportunity for a surface to answer from something other than the shipped truth.
Coverage, unchanged
0.33.1 is matrix-neutral. All seven coverage matrices are unchanged — SOC 2 (AICPA TSC), HIPAA §164.312, NIST CSF 2.0, PCI DSS v4.0.1, ISO/IEC 27001:2022, CIS Controls v8 and GDPR Article 32 infrastructure substrate — and the catalogue is unchanged at 28 Enterprise plugins (27 cloud auditors plus one Zero Trust posture assessment scored from a network-host scan), 55 in total with Community Edition’s 27. One command across all three clouds with every framework enabled evaluates 1,608 controls and returns seven auditor-shaped reports.
Getting it
Community Edition is free and MIT-licensed on npm as nsauditor-ai; the skill package is nsauditor-ai-agent-skill. Enterprise Edition ships as @nsasoft/nsauditor-ai-ee, is licensed per seat, and is also available through AWS Marketplace. Enterprise 0.33.1 requires Community Edition 0.2.37 or newer.
Product details: https://www.nsauditor.com/ai/enterprise/
