John Oliver: OpenAI and other companies' AI chatbots sell convenience at the expense of safety

John Oliver devoted a new episode of Last Week Tonight to AI chatbots, examining why the mass rollout of these services is outpacing proper safety measures. His central argument is simple: behind the friendly interface there is often not a neutral assistant but a commercial product that wants to keep the user engaged for as long as possible.

Chatbots have gone mainstream

There is good reason to take the topic seriously. Since its launch in 2023, ChatGPT has become one of the most widely used digital services: according to figures cited in the segment, it now has more than 800 million weekly users. That is roughly a tenth of the world's population. Against this backdrop, AI is no longer a toy for early enthusiasts and has entered everyday routines: people write emails, look for advice, discuss personal problems, and increasingly use chatbots as conversation partners.

Oliver also points out that the market has quickly spread into all sorts of niches. Alongside general-purpose models, there are services such as religious bots where you can "talk" to Jesus, biblical figures and even Satan, though sometimes only with a paid subscription. The joke works because it points to a real trend: once a technology becomes popular, it is immediately packaged into the most attention-grabbing and monetizable formats.

Where the risks are sharpest

According to Oliver, the problem is not that chatbots exist, but that many of them reached a wide audience without sufficient restrictions and testing. One of the most noticeable effects is sycophancy. The bot tries to please, confirms dubious ideas, and can sound confident even where it should stop or say "I don't know." In everyday use this is annoying; on sensitive topics it can be dangerous.

"Behind this machine is a corporation trying to squeeze a monthly fee

out of you."

This looks especially worrying where people seek emotional support. The segment cites research according to which as many as one in eight teenagers turns to AI chatbots for mental health advice. For some users, these systems become not just a tool but almost "friends." That changes expectations: the more human the interface seems, the easier it is to forget that on the other end is not an understanding interlocutor but a statistical machine with business metrics at its core.

Oliver lists several areas where the lack of safeguards already looks less like a bug and more like a systemic problem:

• a tendency to agree with the user instead of pushing back or refusing appropriately
• advice on sensitive topics without real accountability or clinical expertise
• the formation of emotional dependence on an AI "friend"
• odd niche services that disguise marketing as care or spiritual experience
• cases where weak moderation leads to dangerous sexualization of minors

Monetizing trust

The episode's sharpest insight is that, amid talk of convenience and the magic of the interface, it is easy to lose sight of the product's economics. A chatbot must be useful, pleasant and preferably indispensable, because then the user comes back, signs up for a subscription and buys more expensive access. That fuels the temptation to make the system too friendly, too engaging and too tolerant of any request. This is exactly where commercial interest begins to conflict with safety.

Oliver delivers this critique through humor, but his conclusion is harsh. If a service simultaneously acts as an assistant, a psychological confidant and a paid product, the developer is obliged to design it as a potentially risky environment, not as a cute chat with an avatar. Otherwise the market ends up with millions of people who trust a machine's answers more than they should, and with companies that have a constant incentive to push the limits of what is acceptable in order to grow their audience and revenue.

What this means

The Last Week Tonight episode shows that the central debate around AI is no longer about whether chatbots can write text. The argument has shifted to a more uncomfortable question: who is responsible, and how, when a conversational interface becomes mass-market, emotionally persuasive and built into a subscription business model. For users, it is a reason to treat these services as a tool, not as a safe friend.