OpenAI's Ethics Head Quits, Anthropic Bets $9.1 Billion on Inference, and the White House Calls a Rival 'a Nuisance'

OpenAI's head of ethics departed after less than a year in the role — the latest in a pattern of turnover in safety and ethics leadership at the world's most prominent AI company that sophisticated observers said was a signal about organizational priorities rather than a personnel management footnote. Simultaneously, OpenAI rolled out unlimited free text chats to its user base, its former robotics lead joined Anthropic, and senior White House officials publicly described OpenAI's new head of strategic futures, Dean Ball, as 'a nuisance' over his AI policy commentary — suggesting either a breakdown in a key regulatory relationship or factional disagreement within the administration about AI governance.

Anthropic, meanwhile, made a $9.1 billion data center lease commitment with Riot Platforms — a company better known for Bitcoin mining operations, whose massive electricity infrastructure translates directly to AI compute. The bet reflects a structural shift that research firm Gartner confirmed: for the first time in 2026, AI inference spending has overtaken model training. When the cost of running models exceeds the cost of building them, the technology has entered a production deployment phase that demands different infrastructure, different cost structures, and different organizational capabilities than the research phase that preceded it.

Google's reported moves reinforced how competitive the interface layer of AI has become. The company is reportedly testing an AI-first homepage that drops the classic search button, and separately planning to retire free Gemini Gems on October 20th, pushing users toward paid Spark Skills. Both moves, individually, are competitive product decisions; together, they raise antitrust questions about whether a company dominant in search is using that position to foreclose competition in AI interfaces — the kind of analysis that goes beyond market share alone to examine whether dominance in one market is being leveraged to exclude rivals in another.

Workers building AI systems reported a different reality than the productivity-liberation narrative suggests. Despite promises of shorter work weeks enabled by AI assistance, workers in AI development roles were reportedly logging up to 90 hours per week, according to BBC reporting — a particular irony for people building tools ostensibly designed to reduce human labor burdens.