A public IT company in Rio de Janeiro, Brazil, IplanRIO, announced an AI model called 'Rio 3.5 Open 397B,' but analysis revealed it is likely a mixture of two existing models in a ratio of approximately 6:4. Following criticism, IplanRIO explained they had 'accidentally uploaded an incomplete model.' Nex-AGI, an AI development organization, examined the model and found that when they removed the fixed instruction telling the AI 'You are Rio,' 79% of responses identified as 'Nex from Nex-AGI,' and 0% identified as 'Rio.' The AI also reproduced Nex-AGI's unique organizational introduction. Comparing 'weight tensors' across all 60 layers showed almost all matched values that were a mixture of approximately 60% from Nex-AGI's 'Nex-N2-Pro' and approximately 40% from Alibaba's 'Qwen3.5-397B-A17B.' The degree of agreement was so high that it is difficult to explain with a model that had undergone typical additional training, and no evidence of proprietary training could be found.