ARYA: A Physics-Constrained Composable & Deterministic World Model Architecture
arXiv:2603.21340v1 Announce Type: new Abstract: This paper presents ARYA, a composable, physics-constrained, deterministic world model architecture built on five foundational principles: nano models, composability, causal reasoning, determinism, and architectural AI safety. We demonstrate that ARYA satisfies all canonical world model requirements, including state representation, dynamic prediction, causal and physical awareness, temporal consistency, generalization, learnability, and planning and control. Unlike monolithic foundation models, the ARYA foundation model implements these capabilities through a hierarchical system-of-system-of-systems of specialized nano models, orchestrated by AARA (ARYA Autonomous Research Agent), an always-on cognitive daemon that executes a continuous sense-decide-act-learn loop. The nano model architecture provides linear scaling, sparse activation, selective untraining, and sub-20-second training cycles, resolving the traditional tension between capa
arXiv:2603.21340v1 Announce Type: new Abstract: This paper presents ARYA, a composable, physics-constrained, deterministic world model architecture built on five foundational principles: nano models, composability, causal reasoning, determinism, and architectural AI safety. We demonstrate that ARYA satisfies all canonical world model requirements, including state representation, dynamic prediction, causal and physical awareness, temporal consistency, generalization, learnability, and planning and control. Unlike monolithic foundation models, the ARYA foundation model implements these capabilities through a hierarchical system-of-system-of-systems of specialized nano models, orchestrated by AARA (ARYA Autonomous Research Agent), an always-on cognitive daemon that executes a continuous sense-decide-act-learn loop. The nano model architecture provides linear scaling, sparse activation, selective untraining, and sub-20-second training cycles, resolving the traditional tension between capability and computational efficiency. A central contribution is the Unfireable Safety Kernel: an architecturally immutable safety boundary that cannot be disabled or circumvented by any system component, including its own self-improvement engine. This is not a social or ethical alignment statement; it is a technical framework ensuring human control persists as autonomy increases. Safety is an architectural constraint governing every operation, not a policy layer applied after the fact. We present formal alignment between ARYA's architecture and canonical world model requirements, and report summarizing its state-of-the-art performance across 6 of 9 competitive benchmarks head-to-head with GPT-5.2, Opus 4.6, and V-JEPA-2. All with zero neural network parameters, across seven active industry domain nodes spanning aerospace, pharma manufacturing, oil and gas, smart cities, biotech, defense, and medical devices.
Executive Summary
ARYA, a physics-constrained composable and deterministic world model architecture, is presented in this paper. Built on five foundational principles, ARYA demonstrates satisfaction of all canonical world model requirements and boasts linear scaling, sparse activation, and sub-20-second training cycles. A key contribution is the Unfireable Safety Kernel, ensuring human control persists as autonomy increases. ARYA outperforms state-of-the-art models across six benchmarks, including GPT-5.2 and Opus 4.6, without neural network parameters. This architecture has significant implications for AI safety and control, particularly in high-stakes industries such as aerospace and defense.
Key Points
- ▸ ARYA is a composable, physics-constrained, and deterministic world model architecture
- ▸ ARYA satisfies all canonical world model requirements and boasts improved training efficiency
- ▸ The Unfireable Safety Kernel ensures human control persists as autonomy increases
Merits
Strength in AI Safety
ARYA's Unfireable Safety Kernel provides a technically grounded framework for ensuring human control in AI systems, addressing a critical concern in AI development.
Efficient Training and Scalability
ARYA's hierarchical nano model architecture enables linear scaling, sparse activation, and sub-20-second training cycles, resolving the traditional tension between capability and computational efficiency.
Demerits
Complexity and Implementation Challenges
ARYA's system-of-systems architecture may be complex and challenging to implement, particularly in resource-constrained environments.
Limited Generalizability
ARYA's performance on only six benchmarks, although state-of-the-art, may not generalize to other domains or tasks.
Expert Commentary
The ARYA architecture presents a groundbreaking approach to composable and deterministic world modeling, addressing critical concerns in AI safety and control. While the Unfireable Safety Kernel is a significant contribution, the complexity and implementation challenges of the system-of-systems architecture cannot be overstated. Further research is needed to explore the generalizability of ARYA's performance and to address the scalability and efficiency concerns in resource-constrained environments. Nevertheless, ARYA's potential to improve AI safety and control makes it an essential consideration for AI developers and policymakers.
Recommendations
- ✓ Further research is needed to explore the generalizability of ARYA's performance and to address the scalability and efficiency concerns in resource-constrained environments.
- ✓ ARYA's architecture should be integrated into AI safety and control frameworks, emphasizing the importance of technical frameworks for ensuring human control.
Sources
Original: arXiv - cs.AI
