OpenAI: scrutiny and shake‑up

OpenAI got hit from three directions in two days: European regulators are weighing new rules for ChatGPT, the company has been swapping out major models inside ChatGPT, and San Francisco police arrested a 20-year-old after a Molotov cocktail was thrown at Sam Altman’s home on April 10. (reuters.com) (help.openai.com) (thenextweb.com) The Europe piece starts with a threshold: under the European Union’s Digital Services Act, platforms with more than 45 million monthly users in the bloc can be put into a tougher category with extra audits, transparency duties, and risk controls. Reuters reported on April 10 that the European Commission said it was analyzing ChatGPT’s user data against that line. (reuters.com) The argument is not just about size. It is about what ChatGPT is, because the European Commission is examining whether it should be treated like a large online search engine or platform, which would pull OpenAI into a rulebook built for systems that shape what millions of people see. (reuters.com) At the same time, OpenAI has been changing the product under users’ feet. Its Help Center says that on February 13, 2026, ChatGPT retired GPT-4o, GPT-4.1, GPT-4.1 mini, OpenAI o4-mini, and GPT-5 Instant and GPT-5 Thinking inside ChatGPT, while keeping application programming interface access unchanged. (help.openai.com 1) (help.openai.com 2) That split matters because ChatGPT and the application programming interface are now moving at different speeds. A company can still build software on older models through the application programming interface even after ordinary ChatGPT users stop seeing those names in the app. (help.openai.com 1) (help.openai.com 2) OpenAI’s replacement plan has two layers. The Help Center says GPT-5.3 Instant is the fast general model for paid workspaces, while GPT-5.4 Thinking is the slower model aimed at harder reasoning tasks. (help.openai.com) Then there is the fallback system, which works like a backup generator when demand spikes. OpenAI says GPT-5.4 mini became available in ChatGPT on March 18, 2026, and for many paid users it steps in automatically when GPT-5.4 Thinking rate limits are reached. (help.openai.com) OpenAI also tied those model changes to pricing and access rules for business customers. Its rate card says flexible Business and Enterprise or Education plans use message credits and model-specific limits, and it repeats that the retired ChatGPT models disappeared from the product on February 13 even though the application programming interface stayed available. (help.openai.com 1) (help.openai.com 2) While Europe was weighing regulation and users were adapting to new defaults, police in San Francisco said a suspect threw an incendiary device at Altman’s home on April 10 and later made threats tied to OpenAI’s offices. Reuters, CNBC, and The Next Web all reported that no injuries were reported and an arrest was made the same day. (cnbc.com) (thenextweb.com) (abcnews.go.com) Put together, the week shows three different kinds of pressure landing on one company at once. Brussels is asking whether ChatGPT is now big enough to regulate like a mass-distribution system, OpenAI is forcing users onto a new model map with automatic fallbacks, and the company’s highest-profile executive just became the target of a real-world attack. (reuters.com) (help.openai.com) (thenextweb.com)