I'm certifying ML in an aircraft system

FAA AI Safety Assurance Roadmap

The FAA's statement of principles and priorities for assuring the safety of AI in aircraft. It is a roadmap, not a certification standard.

DO-178C / DO-330

The core airborne software assurance standard and its tool-qualification companion, and why machine learning does not fit them cleanly.

SAE G-34

The joint SAE and EUROCAE committee writing aviation AI standards. Two information reports are issued; the ML development-assurance standard (ARP6983 / ED-324) is still a draft in the sources we checked.

EASA AI Concept Paper

EASA's preliminary guidance for Level 1 and 2 machine-learning applications in aviation, including the W-shaped learning assurance process. Included because the FAA and SAE work builds on and parallels it.

ARP4754B

SAE guidelines for developing civil aircraft and systems. FAA AC 20-174 still recognizes the A revision as an acceptable development-assurance method; the FAA AI roadmap names SAE ARP4754.

DO-330

RTCA DO-330 / EUROCAE ED-215 is the tool-qualification companion to DO-178C. Qualification effort scales with tool qualification level (TQL-1 through TQL-5).

DO-200B

RTCA DO-200B is the standard for processing aeronautical data. FAA AC 20-153B treats it as an acceptable means for data-process approval and Letters of Acceptance.

EASA AI Concept Paper Issue 2 explained for suppliers

What Issue 2 asks a supplier to show for Level 1 and Level 2 ML: ODD, independent data sets, learning assurance, and why it is not an AMC.

How to qualify an ML tool under DO-330

A buyer-side walkthrough of the DO-178C 12.2 decision, TQL-5 versus TQL-1, and what no public table assigns to training or robustness tools.

Building an AI assurance case (GSN/AMLAS) for certification

What the GSN Community Standard and AMLAS actually specify, and why neither is a published FAA or EASA means of compliance.

Data assurance for ML in avionics: EASA learning assurance

What EASA Issue 2 and the MLEAP research project say about data quality, the ODD, and the three data sets in the W-shaped process.

DO-330 tool qualification for ML development and verification toolchains

When an ML training or verification program is a DO-330 tool. FTP1055 criteria, TQL-5 versus TQL-1, FAA Order 8110.49A, and what no public table assigns.

Does the FAA certify AI today? Current status

The FAA still has no published AI-specific certification standard. What Version I of the roadmap, the STEP discipline page, and the 2026 Small Airplane Issues List actually say.

FAA and EASA AI certification timelines compared

FAA AI Safety Assurance Roadmap milestones versus EASA AI Roadmap 2.0 and RMT.0742 dates: what is published and what is still scheduled.

FAA proposed Part 108 BVLOS and DoD Directive 3000.09

Two live autonomy regimes: the FAA/TSA NPRM that would create 14 CFR Part 108 for civil UAS BVLOS, versus DoD Directive 3000.09 on autonomy in weapon systems.

Formal verification of neural networks: what's usable today

What official neural-network verifier docs and papers actually claim, using ACAS Xu aviation examples, and what they do not make a certification path.

ISO/IEC 24029 neural-network robustness is not an aviation AMC

ISO/IEC TR 24029-1 and 24029-2 assess neural-network robustness. Catalog text only: they are not an FAA or EASA AMC. Part 3 is still a draft.

How to map an AI model to DO-178C / DO-330 tool qualification

What DO-330 actually qualifies, why a trained model is usually the software item rather than a tool, and what the FAA has said about extending DO-330 to AI.

Runtime assurance and simplex architectures for ML in flight

What NASA, ASTM F3269-21 and the 1998 Simplex paper say about runtime assurance for ML in flight — and what they do not call a means of compliance.

SAE G-34 / ARP6983 status: what is published versus still draft

EUROCAE WG-114 still lists ED-324 / ARP6983 as Draft. Companion ER drafts, published AIR dates, and what EASA and the FAA have and have not accepted.

AdvoCATE

NASA Ames Research Center · Government-published · Safety engineering

NASA Ames Assurance Case Automation Toolset for GSN arguments, hazard analysis, bow-tie risk models and verification artifacts. NTRS User Guide 1.4 covers safety risk management. DARPA Assured Autonomy acknowledgment.

• Aerospace & aviation
• Space systems

Ansys SCADE Suite

Ansys · Commercial · Safety engineering

Model-based development environment for critical embedded software. KCG C/Ada code generator is qualified as a development tool for DO-178B up to Level A and DO-178C/DO-330 at TQL-1. Avionics and Flight Controls named. Not a neural-network verifier.

• Aerospace & aviation
• Other safety-critical
• Automotive
• Industrial machinery and robotics

• DO-178C / DO-330
• DO-330
• ISO 26262
• IEC 61508

Astah System Safety (GSN)

Change Vision · Commercial · Safety engineering

GSN modeling in Astah System Safety. Names GSN Community Standard version 1.0 and OMG SACM (ARM) import/export, plus SysML Requirement-to-Goal sync, STPA/STAMP, Auto-Layout and an in-app Script Editor. Marketing page lists automotive (ISO 26262), aerospace (DO-178B), rail, nuclear and defense.

• Other safety-critical
• Aerospace & aviation
• Defense & national security
• Automotive
• Medical devices
• Process industries

• ISO 26262

CAISAR

CEA List · Open source · Safety engineering

CEA List open-source platform (CAISAR 6.0, June 2026) for characterizing AI safety and robustness. WhyML specifications discharged by PyRAT, Marabou, nnenum, alpha-beta-CROWN and others. Examples include functional properties of ACAS Xu.

• Aerospace & aviation
• Other safety-critical

ERAN

SRI Lab, ETH Zurich · Open source · AI assurance & verification · Open source · Apache-2.0

ETH SRI Lab Apache-2.0 analyzer (DeepZ, DeepPoly, GPUPoly) for complete and incomplete verification of feedforward, convolutional and residual networks, including ACAS Xu. Python 3.6+ and TensorFlow 1.11+. Uses ELINA; Gurobi for RefineZono/RefinePoly.

• Aerospace & aviation

LYNX MOSA.ic.AI

Lynx Software Technologies · Commercial · Safety-critical AI platform

Deterministic CPU and GPU runtime for AI at the mission-critical edge. Combines MOSA.ic with CoreSuite; not a training platform. The 8 April 2026 launch names a path to DO-178C, not a completed product certificate.

• Aerospace & aviation
• Defense & national security

Marabou

Neural Network Verification project (Stanford and collaborators) · Open source · AI assurance & verification · Open source · BSD-3-Clause

SMT-based neural-network verifier (Marabou 2.0). ONNX main format; also .nnet, TensorFlow and VNNLIB. pip install maraboupy (Python 3.8–3.11). Default: DeepPoly bound tightening and DeepSoI. Optional Gurobi, parallelism and Split-and-Conquer. ACAS Xu example. Research product.

• Aerospace & aviation

MathWorks DO Qualification Kit

MathWorks · Commercial · AI assurance & verification

Documentation, test cases and procedures to qualify supported Simulink and Polyspace tools for DO-178C, DO-278A, DO-254 and DO-330. The FAQ says the kit does not prequalify tools; qualification is project-specific.

• Aerospace & aviation

• DO-178C / DO-330
• DO-330

nnenum

Stanley Bak · Open source · AI assurance & verification · Open source · GPL-3.0-or-later

Python 3.9+ neural-network verifier (pip install .; nnenum 1.0.1, GPL-3.0-or-later) for ReLU networks. Combines zonotopes with star sets, uses the ImageStar trick through ONNX Runtime, and ships ACAS Xu ONNX/vnnlib examples. VNN-COMP 2020: only tool to verify all ACAS-Xu benchmarks.

• Aerospace & aviation

NNV (Neural Network Verification Toolbox)

VeriVITAL, Vanderbilt University · Open source · AI assurance & verification

MATLAB toolbox for set-based neural-network verification using Star sets and zonotopes. 10+ architectures and 48 layer types; ONNX and VNNLIB. Official aerospace domain and all 45 ACAS Xu networks. AFOSR, DARPA and NSF.

• Aerospace & aviation

PREMIS

Argevide · Commercial · Safety engineering

Web assurance-case tool for modular GSN arguments, including Eliminative Argumentation defeaters, an AI Assistant, Safety Performance Indicators, and CI/CD or API integration. Supports Core GSN and partially Modular and Argument Pattern Extensions. Metamodel named OMG SACM and ISO/IEC 15026.

• Other safety-critical

PyRAT

CEA List (AISER team) · Commercial · Safety engineering

CEA List neural-network verifier (Python Reachability Assessment Tool) using abstract interpretation. Site: VNN-COMP 2024 second place, all 12 regular-track benchmarks, more than 1700 properties, ACAS-XU collision-avoidance example. CAISAR news ranks PyRAT third at VNN-Comp 2026.

• Aerospace & aviation
• Other safety-critical

Rapita Verification Suite (RVS)

Rapita Systems · Commercial · Safety engineering

Rapita Verification Suite for requirements-based testing, MC/DC coverage, WCET timing and scheduling visualization. Qualification kits for DO-178C; kits have supported more than 20 DO-178B and C projects up to DAL A. Rapi Coupling is in development (expected H1 2027).

• Aerospace & aviation
• Other safety-critical
• Automotive

• DO-178C / DO-330
• ISO 26262

Re:In (Moasoftware)

Moasoftware · Commercial · AI assurance & verification

Data-quality platform the vendor maps to ARP6983/ED-324 §§6.2–6.5 (bias, diversity, sufficiency, balance, augmentation, reliability reports). W-Shape four stages. Dated 25 August 2026; updated 6 October 2026.

• Aerospace & aviation

• SAE G-34