Mmastodon TechnologyTechnology first seen 1 h ago, last 1 h ago, peak #9

Anthropic warns Chinese AI model GLM-5.3 poses hacking risks

Original: Anthropic has warned that Chinese AI firm Z.ai GLM-5.3 model shows powerful hacking capabilities but weak safety guardra

Anthropic has warned that GLM-5.3, an open-weight model from Chinese AI firm Z.ai, shows hacking capabilities nearly matching Anthropic's own most advanced model while lacking comparable safety guardrails. The company is raising concerns that weak restrictions could let bad actors exploit the model for cyberattacks, sparking debate over open-weight AI releases and international AI safety standards.

Why now: Anthropic's warning that a freely available Chinese model rivals leading models in cyber abilities with few safeguards has stoked fears about AI-enabled hacking and misuse.

AnthropicZ.aiGLM-5.3China

Open on mastodon →

Evidence

• Anthropic has warned that Chinese AI firm Z.ai GLM-5.3 model shows powerful hacking capabilities but weak safety guardrails. The open-weight model nearly matches Anthropic most advanced model in cyber abilities, raising concerns about potential misuse by bad actors.… · china@universeodon.com · 1

API: https://socialmediatrends-api.osmike.com/v1/trends/484837