Corrosion inhibitor performance in high-pressure, high-temperature (HPHT) oilfield environments is governed by coupled electrochemical kinetics, competitive adsorption, and thermally induced molecular degradation, which are rarely captured simultaneously in conventional evaluation frameworks. This study develops a hybrid physics-informed machine learning (PIML) architecture to quantitatively assess inhibitor stability under extreme thermodynamic conditions (T >150 °C, P >50 MPa, high salinity brines). The framework embeds Langmuir–Temkin adsorption isotherms, Arrhenius-type degradation kinetics, and Butler–Volmer electrochemical relationships into a neural network structure through physics-based regularization constraints. Experimental datasets comprising electrochemical impedance spectroscopy (EIS), potentiodynamic polarization curves, and mass loss measurements are integrated with thermodynamic state variables and fluid chemistry descriptors. Feature representations explicitly account for inhibitor molecular structure (e.g., functional groups, polarity indices) and competitive adsorption with CO?/H?S species. The model predicts time-dependent inhibition efficiency, surface coverage evolution, and desorption thresholds under transient flow and thermal cycling conditions. Validation against unseen HPHT datasets demonstrates reduced extrapolation error and improved mechanistic consistency relative to purely data-driven models. Sensitivity analysis identifies salinity-induced double-layer compression and temperature-driven desorption as dominant destabilization pathways. The proposed framework enables targeted inhibitor formulation design and adaptive dosing strategies for harsh oilfield environments.