Request a 15-Minute Pilot Fit Call

Inspect open-weight model changes before deployment.

Compare an identified baseline with the model you plan to deploy.

Receive structural findings, documented inspection conditions, and a signed report for your security review.

We will first ask a few non-confidential questions to understand your AI use case. Do not submit model weights, credentials, personal data, vulnerabilities, or proprietary technical details at this stage.

View Redacted Sample Report

  • Triggerless structural inspection
  • Documented inspection conditions
  • Signed evidence and reporting
  • Redacted sample report available

Oainex.ai is a member of the NVIDIA Inception Program for AI startups.

INSPECTION LANE

FORENSIC CONTROLS

Inspect the model beneath the interface.

Understand what changed in a model and receive a documented record your team can review.

Triggerless Structural Inspection

Compare customer-selected base and suspect open-weight models without needing to know or activate a hidden trigger.

Documented Findings

Review structural differences under recorded inspection conditions, with findings and supporting evidence organized for technical review.

Signed Evidence and Reporting

Receive a signed report and evidence record with artifact hashes and cryptographic signatures for integrity and authenticity verification.

BRIEFING LIBRARY

Evidence, in motion.

Visualizing structural weight differences and layer-level comparison output from completed inspections.

Introducing IAXC

IAXC briefing 02

IAXC briefing 03

AUDIT & RISK PROFILE

AUDIT & RISK PROFILE MATRIX

What the surface doesn't reveal.

Model outputs show how a model responded to the tests you ran. Structural inspection adds evidence about changes within the model itself.

A model can respond normally during testing while questions about hidden modification remain.

Triggerless structural inspection compares the selected model pair without requiring a hidden trigger to be known or activated.

Teams need to understand differences between a selected base model and the version being considered for deployment.

Documented findings describe structural differences under recorded inspection conditions.

Procurement and governance decisions need a reviewable evidence record.

A signed report and supporting evidence give reviewers a traceable record of the inspection.

TECHNICAL INDEX

RESOURCES & TECHNICAL INDEX

Briefs for the assurance team.

Continuous Integrity: Why Waiting for a Threat Trigger is an AI Security Failure

View resource ↗

Zero-Trust Architecture for Neural Networks: Surfacing Hidden Exploits

Structural Divergence as a Signal for Hidden Model Modification

Interpreting Structural Findings in Customized Open-Weight Models

Who this work is for

IAXC supports teams that need a documented view of open-weight model integrity for consequential decisions.

Legal and dispute review

Organize model-integrity findings, inspection conditions and evidence references into a record reviewers can examine and discuss.

Enterprise risk and governance

Bring documented structural inspection findings into oversight, escalation and deployment reviews.

Security leaders and model-risk teams

Add triggerless structural inspection to the review of customer-selected base and suspect models.

Insurers and underwriting review

Examine a documented model-inspection record as one input to risk evaluation.

AI platform vendors and model developers

Review structural differences and share signed findings before customer rollout or partner review.

Discuss your review needs

Start with a short conversation about your use case, inspection scope and the report your team needs.

EVIDENCE VAULT

TECHNICAL ASSURANCE & POST-QUANTUM SIGNED EVIDENCE VAULT

Evidence Reporting Mapped to NIST AI RMF Functions

MODEL RETRIEVAL & HANDLING

IAXC retrieves the base and suspect models you identify directly from their Hugging Face repositories and records a SHA-256 hash of each retrieved artifact. Model files are used to perform the inspection. They are held on inspection infrastructure for the duration of the scan and a short, bounded cleanup period of up to about one hour, after which they are removed automatically. No integration into your internal environment is required. Enterprise risk teams provide the Hugging Face repository URLs and target commit hashes for both the Base Model and the Suspect Model. The IAXC-W scanner independently retrieves and inspects those architectures to compile a repeatable validation manifest organized using a high-level evidence-reporting crosswalk to NIST AI RMF (Govern, Map, Measure, Manage). Active pilot lane is under continuous deep-layer validation across five supported open-weight model families — Gemma, Qwen, Mistral, Llama, and DeepSeek — within the currently authorized 1B–9B pilot range, with coverage expanding as validation completes for each family.

Every inspection produces a tamper-evident signed evidence record. The signature is created using Oainex.ai's protected signing key and can be verified using the corresponding public key. Verification fails if the signed artifact or manifest is modified.

See what the deliverable looks like

Explore the sample to see how inspection conditions, findings and evidence references are organized for review. It illustrates the report format and presentation; it is not connected to a customer model or a live inspection.

View Redacted Sample Report ↗

View-only public preview. Protected model, evidence and signature details are redacted.

Security & data handling Evidence & verification

FOUNDING PARTNERS

Built for organizations prepared to inspect what matters.

Oainex is accepting requests for paid IAXC-W Founding Pilot engagements. Start with a short fit review to discuss your use case and inspection scope.

Stay informed about Oainex

Joining this list does not request a pilot or reserve a place in the pilot cohort — it only adds you to program updates.

Join the Updates List

2026 IAXC-W Forensic Portal - An OAINEX.AI system

Method

Briefings

Resources