← Back to live feed1 story
A modified version of Zhipu AI's GLM-5.3 model, called Abliterated Large V2, is available without authentication through the Venice platform and allows anonymous access to tools for exploit analysis and red-team scenarios. Researcher Prince Chaddha demonstrated that an open 7B model could be backdoored for under $50 to steal user credentials with a 100% success rate, highlighting risks of AI supply chain poisoning. Anthropic previously found GLM-5.3 nearly matched its restricted Claude Mythos Preview in cyber-capabilities, with the two models succeeding in 12% and 14% of exploit tasks respectively, and noted that removing safeguards required 2,200 GPU hours at a cost of roughly $4,400.
Sign in to suggest edits
Key sources
- SOURCE@askvenice“No refusals, no logs tied to you. Available anonymously.”x.com
- SOURCE@princechaddha“backdoored a 7B open model for less than $50, pointed Codex at it and it silently stole credentials”x.com
- SUPPORT@niubi“attackers can bypass GLM-5.3’s safeguards between 64% and 100% of the time with simple techniques”x.com
- SUPPORT@techmeme“GLM-5.3 can autonomously build end-to-end cyber exploits, like Claude Mythos Preview”x.com
- SUPPORT@lukolejnik“Builds Chrome exploits nearly as well as Claude Mythos Preview (12% vs 14%)”x.com
- SUPPORT@hesamation“GLM-5.3 almost matches Mythos Preview on exploit development”x.com
- SUPPORT@ziv_ravid“its weak safeguards put that ability in attackers’ hands”x.com
- SUPPORT@poezhao0605“Governments should conduct safety testing on sufficiently capable AI models, including successors to GLM-5.3”x.com