Independent AI Certification Authority
Trust claims in AI
should be verifiable,
not asserted.
Fidensa independently certifies AI capabilities — MCP servers, Skills, plugins, and workflows — through behavioral testing, adversarial analysis, and supply chain verification. Every certification is a signed, portable artifact that anyone can verify against the evidence.
The structural problem
Today, every trust claim in the AI ecosystem is self-asserted. When a vendor says “this is safe,” that’s marketing. When a government says “this vendor is a risk,” that’s politics. When a publisher says “this capability is reliable,” there is no independent evidence to verify it.
Every dispute degrades into “who do you believe?” because there is no neutral, evidence-based infrastructure for resolving it. No independent body examines AI capabilities and produces signed, testable, portable artifacts documenting what was found.
The structural answer
The AI ecosystem needs the same kind of institution that the electrical products industry built over a century ago. An independent authority that doesn’t manufacture, sell, or regulate — it tests and certifies. Its entire existence depends on the integrity of its process.
Independent
Not owned by or beholden to any AI vendor, platform, or government.
Evidence-based
Every certification backed by behavioral testing, supply chain analysis, adversarial testing, and ongoing monitoring.
Verifiable
Signed, checksummed, portable artifacts. Anyone can verify against the authoritative record.
Neutral
The same process applies regardless of which vendor built it or which platform distributes it.
Six-stage verification
Every capability passes through a defined pipeline. Each stage produces artifacts that feed the trust score and are published in the contract.
Ingest
Source acquisition, build, interface extraction, provenance hash
SBOM & Supply Chain
Dependency tree mapping, vulnerability cross-reference (syft, grype, osv-scanner)
Security Scan
Static and behavioral analysis (Cisco mcp-scanner, skill-scanner)
Functional Test
Valid, edge, error, and LLM-generated test cases in isolated sandbox
Adversarial Test
Structured attack library — 6 categories, 40+ patterns — with finding severity evaluation
Behavioral Fingerprint
Per-tool timing (p50/p95/p99), error rates, resource profiling, baseline for drift detection
Stage 4 (Certify) assembles the contract, computes the trust score from eight weighted signals, signs the artifact with ES256 (ECDSA P-256), and issues the certification.
Live certifications
Batch 1 — three capabilities evaluated through the full pipeline. Real data, real findings, real scores. Every attestation is queryable right now.
mcp-server-filesystem
MCP Server · v0.6.3 · Model Context Protocol a Series of LF Projects, LLC.
◉ Verified
67/D
docx-skill
Skill · vlatest · anthropics
△ Evaluated
43/F
mcp-server-everything
MCP Server · v2.0.0 · Model Context Protocol a Series of LF Projects, LLC.
✓ Certified
68/D
Runtime attestation (try it)
GET https://fidensa.com/v1/attestation/mcp-server-everythingCertification tiers
Score and certification are separate. The trust score is an analytical summary. The tier is a judgment based on finding-severity thresholds.
Fidensa Certified
No unmitigated critical findings. No more than two unmitigated high findings. Pipeline completed.
Fidensa Verified
Pipeline completed. Findings of any severity documented. Evidence-backed contract issued.
Fidensa Evaluated
Pipeline ran with partial coverage. Incomplete data. Contract documents what was found.