“General-purpose AI systems have persistent reliability challenges, demonstrating high performance on some tasks while failing completely on others, as noted in the International AI Safety Report 2025.”