Aerospace and Defence and Artificial Intelligence: Assurance Before Autonomy, Space as an Early Proving Ground, and Military Use Governed Outside the Civil Rules

Abstract

Purpose. Aerospace is where automation has been trusted with human lives for longest, and where the standards for proving a system safe are the strictest. This article asks where artificial intelligence stands in aerospace and defence in 2026 and where it may be by about 2035. Design and methods. A review combining the share of the aerospace and defence research literature that involves AI (OpenAlex, 2010–2025), the guidance of the European Union Aviation Safety Agency and the US Federal Aviation Administration on AI in aircraft, the first on-board AI satellite mission, a publicly reported test of AI agents flying a fighter against a human pilot, military and international policy on autonomy in weapons, and the EU AI Act. Claims taken from the literature were checked against the sources cited. The scenarios are judgements built on this evidence. Findings. The share of aerospace and defence works that mention AI rose from 2.8% in 2018 to 12.9% in 2025, 4.6 times as much, with AI-related works growing 31.9% a year, but only 5.0% of these AI works involve language models, one of the lowest shares in our series, so most of the field is perception, control and optimisation, not generative AI. The bottleneck is assurance. European and US aviation authorities describe a staged approach, with EASA's guidance moving from AI that assists humans, through human–AI teaming, to a proposed third level of advanced automation released for consultation in June 2026, and the FAA separating AI whose parameters are fixed before deployment from AI that keeps learning in service. Space has been an early proving ground: a European satellite launched in 2020 ran a neural network on board to discard cloudy images before downlinking them. In defence, a publicly reported 2023 test flew AI agents in a modified fighter against a human pilot, a US directive requires senior review of autonomous weapon systems, and in December 2024 the United Nations General Assembly adopted a resolution on autonomous weapons by 166 votes to 3. The EU AI Act excludes systems used exclusively for military purposes. We expect AI to remain outside the control loop of passenger aircraft through 2028, to spread in design, maintenance, operations and satellite data processing, and the first certifications of learning-based functions in civil aviation to come in 2029–2032, probably for functions that advise the crew. Originality. The article treats certification and governance, not capability, as the pace-setting factor in aerospace, separates what is publicly documented in defence from what is not, and draws on the regulators' own staged frameworks.