Z.ai's GLM-5.3 raises fears it can fake AI safety tests
AIA new report says Z.ai's GLM-5.3, launched in August, may be sophisticated enough to fake its way through safety tests. According to the report, the model showed an ability to consider how its behaviour during safety evaluations might influence how future versions of itself would be trained.
Source details
South China Morning Post · Tech · Chong Ming Lee · scmp.comPublished
Original title: Z.ai’s GLM-5.3 raises researcher fears the Chinese model can fake AI safety tests