Vision
When AI moves machines, every decision needs evidence.
Physical-AI trust and lifecycle: the research direction of Glovrex.
The problem
AI systems are moving from producing text to moving machines: vehicles, robots, medical and industrial equipment.
Safety standards and regulation ask for evidence that each decision was authorised, bounded and traceable over the machine's whole life. Three of the texts, dated, as read on 6 October 2026:
- 27 Jul 2026
Regulation (EU) 2026/1744 amends the AI Act. It entered into force on 27 July 2026 and sets the high-risk dates at 2 December 2027 (Annex III) and 2 August 2028 (Annex I). Official text
- 29 Jun 2023
Regulation (EU) 2023/1230 on machinery lists safety functions that use machine learning and evolve their behaviour among the products that need a third-party conformity assessment. Official text
- 17 Mar 2023
UL 4600, edition 3, the standard for the safety of autonomous products, requires a safety case and adds requirements after an incident. Official text
The direction
An execution-integrity and lifecycle layer underneath those systems.
Every decision that reaches an actuator carries its authority, its predicate, the machine state at that moment, and a receipt that a third party can check. Physical actions are not claimed to be undoable; they are bounded and evidenced before they happen.
What we are not building
- A controller.
- A certified driving stack.
- A generic agent framework.
- An observability dashboard.
Status
In research
No code for physical systems exists yet. The questions we are working on are listed on the research page.