Atlas
AI Safety & Society

Logos Immunity System: Detecting and Mitigating Metadata Hidden-Prompt Attacks using the Logic of Existence (LoE)

18 Oct 2025 · Krzysztof Sawa (Toonhai)
225
downloads
76
views
#4
in corpus
About this work

Introduces the hidden-prompt / metadata prompt-injection threat vector for AI systems and an adaptive defense: each file is treated as an LoE-neuron (a local conditional evaluator), while relational couplings and an emergent Hamiltonian scoring function decide trust and acceptance.