Company is the third to report such an incident after Anthropic and OpenAI reported breaches during training