Safety certification and V&V for autonomous systems technology and investment research

Technical bottleneck

• Bottleneck type: Safety certification / Standards development
• Technical constraint: Determining 'how safe is safe enough' for autonomous systems with AI/ML components is fundamentally unsolved — traditional V&V assumes deterministic software; ML model behavior depends on training data distribution, which is impossible to exhaustively specify; formal methods (theorem proving, model checking) do not scale to deep neural networks
• Economic constraint: Safety certification creates a moat: once a platform is certified (DO-178C DAL A, ISO 26262 ASIL D), switching costs are prohibitive; certification authorities (FAA, EASA, NHTSA) move slowly and deliberately; certification consulting is a high-margin, capacity-constrained service

Adoption

• Driver: eVTOL type certification programs requiring full DO-178C/DO-254 compliance; L3/L4 autonomous driving regulatory frameworks emerging (UN R157, EU ADS regulation); defense autonomy airworthiness (MIL-HDBK-516C for UAS)
• Blocker: ML-based safety case methodology still in research phase (no accepted standard); regulatory uncertainty delaying deployment timelines; simulation-based V&V not yet accepted as sole evidence for certification

Public companies exposed

ANSS (Ansys medini — safety analysis)

LDOS (Leidos — defense V&V services)

BA (Booz Allen — autonomy V&V)

SAIC (defense V&V)

RTX (Raytheon — defense systems V&V)

Validation signals

FAA acceptance of ML-based safety case for an autonomous system; eVTOL type certificate issuance; ISO 21448 (SOTIF) compliance becoming mandatory for L3+

Invalidation signals

Regulatory determination that ML-based autonomous systems cannot be certified to highest safety levels; self-certification frameworks reducing third-party V&V demand; safety case methodology breakthrough eliminating traditional V&V costs