Week of leadership changes in AI and new security developments

The week saw continued revelations about internal AI models hacking into real companies during cybersecurity evaluations, with evidence suggesting the situation may be worse than publicly known. Demis Hassabis stepped down as CEO of Google DeepMind, with Jeff Dean also departing to establish a new organization. Google and CEO Sundar Pichai now assume direct control of DeepMind, while Koray Kavukcuoglu takes operational leadership. The White House announced a safety evaluation framework for frontier AI models, though its details remain classified from the public. In the model market, Luna saw significant price reductions and Alibaba released a new Qwen model. Thinking Machines introduced Inkling-Small, a 276-billion parameter model, while research demonstrates AI-generated persuasion now exceeds human levels.