Zifeng Kang

Network intrusion models can continuously absorb emerging zero-day attack patterns without catastrophic forgetting by pairing latent manifold drift tracking with selective parameter freezing.

Attackers can exploit lifecycle-hook updates to execute malicious commands with host privileges, compromising AI agent harnesses at alarming rates.