This invention relates to the field of
data security technology and proposes an intelligent
password management and secure login method under dynamic credential control. The steps include: creating a high-strength
password using a built-in
password generator; dynamically calculating an account security
trust score based on a neural
network model trained from multi-dimensional
behavioral data; and triggering a risk-based targeted password update when the
trust score is too low. During the login process, a pre-built user behavior
knowledge graph and a heterogeneous graph
attention network model are used to calculate the context deviation of login events in real time, identify complex
attack behaviors, and execute tiered protection responses. After the account is locked, a real-time risk
score is initialized and dynamically adjusted based on multi-source risk signals and a nonlinear decay model to achieve intelligent risk unlocking. This invention realizes a transformation of
credential management from static rules to dynamic
cognition, significantly improving the defense capabilities against advanced threats such as credential stuffing and credential collision attacks, while optimizing user experience and operational efficiency.