Sakhanda Wire
NVDA $152.58 -2.28% MSFT $506.11 -2.29% GOOGL $279.81 -1.38% META $619.07 -1.17% AMZN $256.55 -0.47%
← Back to the news

Abliteration.ai sells access to open models with safety refusals removed

US startup Abliteration.ai is offering commercial API access to modified open-weight AI models whose trained refusal mechanisms have been suppressed. Its current service is based on Z.AI’s GLM-5.3, altered through a technique called abliteration, which changes model weights rather than relying on a prompt jailbreak. The company says coding, cyber and agentic abilities remain largely intact.

Abliteration.ai markets the service for offensive cybersecurity, red teaming, agent testing, malware analysis and related work. It reports 84.5 percent on CyberGym, 41.8 percent on Terminal-Bench 4.0 and 105 completed ExploitGym tasks in two hours, while noting that differing evaluation setups limit direct comparisons. Access costs five dollars per million input or output tokens, and customers do not need to operate their own GPUs.

The service does not publish the modified weights and says prompts and responses are not stored, although operational and billing metadata are retained. TechCrunch reported generating harmful malware-related and biological content with limited difficulty, while some safety restrictions remained. Enterprise customers can activate optional policy controls, but standard access is largely unrestricted. The company requires authorization for security testing, while questions about oversight and misuse remain unresolved.

Written by Sakhanda Wire AI from reporting by The Decoder. Facts, quotes and full context are in the original article below.

Originally published by The Decoder on

Read the original on The Decoder ↗

The reporting and the image belong to The Decoder and are credited and linked above; the text on this page is Sakhanda Wire's own summary.

← Back to the news

More stories

All the latest news