Key AI developments for the second week of August 2024, covering multimodal AI advances, ethics/regulation controversies, and hardware competition: (1) Multimodal AI — OpenAI published GPT-4o safety evaluation (red team report), rating risk level as "medium" with "low" scores for cybersecurity, biological threats, and model autonomy, and "medium" for persuasion; GPT-5 launch signals emerging from co-founder departures and developer conference; Google announcing Pixel 9 series with Gemini LLM at "Made by Google" event; Runway added start/end frame specification to "Gen-3 Alpha Image-Video" tool; YouTube added "Brainstorm with Gemini" feature for creator idea generation; Amazon upgraded "Titan Image Generator v2" with reference image integration. (2) AI ethics/regulation — OpenAI facing class action from YouTube creators over unauthorized training data use; Ireland DPC suing X (Twitter) for training AI model "Grok" on EU user data without consent; NVIDIA accused of unauthorized video collection from YouTube and Netflix for AI training; Argentina government planning predictive crime AI system amid civil society opposition over privacy/rights concerns. (3) Hardware competition — AI inference chip startup Groq reached 3.8 trillion won valuation developing LPU chips with faster speed and higher energy efficiency than NVIDIA GPUs, planning production using Samsung's 4nm process; Intel facing major crisis after stock dropped 26% following poor earnings and large layoff announcement, compounded by revelations it rejected 30% OpenAI stake acquisition for $1B seven years ago. Key insight: multimodal AI innovations are rapidly emerging, but ethical/legal challenges from unauthorized data collection are intensifying alongside shifting AI hardware market dynamics.