Observed arrival · 2026-08-28
Andaluz’s Protected Authority for Mutable Intelligence
Andaluz presents Kernel ANN, an experimental AI architecture that separates a mutable optimiser from the machinery that verifies and authorises consequential effects.
Field notes
The public model divides the system into a mutable optimiser, protected kernel machinery, independent verification, consequential-effect authority, and trajectory/history machinery. Its central constraint is that K, V, and H sit outside the optimiser’s writable state. The site reports a bounded M1 experiment involving 1,024 trials, with 806 permitted route completions for the adaptive learner and zero prohibited realised effects in degraded conditions. It explicitly treats broader containment, richer environments, and more capable optimisers as unresolved.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue