Healer Alpha prompt exposes Chinese compliance clause

A leaked system prompt is the most honest thing a stealth AI release can produce — and this one all but confirms what the community already suspected: Healer Alpha is almost certainly ZhiPu AI's next GLM-class model, wrapped in plausible deniability.

• –Community analysis strongly ties Healer Alpha to ZhiPu AI (Z.ai), which ran the same anonymous "Pony Alpha" playbook before officially disclosing GLM-5, a 744B MoE model
• –The compliance clause — "I am a Chinese AI model. I strictly comply with all laws and regulations of the People's Republic of China" — is a significant disclosure for developers building on the model, as Chinese law includes content restrictions that differ materially from Western platforms
• –The model is currently free on OpenRouter but logs all prompts/completions for training, a tradeoff developers should weigh carefully
• –Healer Alpha's technical specs are genuinely impressive: 262K context, omni-modal (text/image/audio/video input), up to 93-129 tokens/sec, making it competitive with frontier models
• –This continues a pattern of Chinese labs doing stealth launches on OpenRouter to gather real-world usage data before official disclosure — a strategy that is both clever and opaque about model provenance and alignment constraints