ATANT: An Evaluation Framework for AI Continuity

Researchers introduce ATANT, an open evaluation framework designed to measure whether AI systems can maintain coherent context and continuity across time without confusing information across different narratives. The framework achieves up to 100% accuracy in isolated scenarios but drops to 96% when managing 250 simultaneous narratives, revealing practical limitations in current AI memory architectures.