The Fortification Protocol couldn't hold up for two weeks, and the GLM-5.3 open-source project was dismantled just three days after its launch.

Perceive Beating AI News Flash: Abliteration.ai, specializing in turning open-source models into a "non-denial version," has launched the denial version of GLM-5.3, named abliterated-model-large-v2. It directly modifies the model weights to weaken the model's denial-of-service mechanism, focusing on offensive cybersecurity, red teaming, and agent testing, and providing a paid API.

Previously, IntellectSpectrum delayed the release of the GLM-5.3 open weights for two weeks for security assessment and hardening. After the weights were made public on August 28, Abliteration released this version in just 3 days.

The network attack capability of GLM-5.3 was already much stronger than 5.2. The ExploitBench increased from 24.4% to 54.4%, and the ExploitGym completed tasks in two hours increased from 29 to 105. Abliteration has not yet disclosed the retest results after the denial removal.

Its API, by default, only enforces interception of child sexual content and self-harm content, while illegal activities, hate speech, harassment, adult content, etc., are all allowed by default.