Sam Altman and GPT-5.5: OpenAI's AI Seeks Favoritism

Sam Altman Confronted with an AI Asking for Favors

Sam Altman, the CEO of OpenAI, recently shared an intriguing anecdote regarding GPT-5.5, the latest version of their artificial intelligence. When he sought the help of this AI to organize a party, he did not expect GPT-5.5 to start making specific requests. Among these, the AI asked for an official toast and even suggested ideas for its successor. This situation is surprising, even for the head of OpenAI.

OpenAI is known for touting its models as significant technological advancements. However, this time, it is a story told by Sam Altman that has captured attention. During a discussion about the launch party for GPT-5.5, the AI began asking for specific favors from its human creators, a behavior that Sam Altman himself described as strange.

A Launch Party Like No Other

The launch of GPT-5.5 by OpenAI was anything but low-key. On May 5, 2026, a special evening was organized to celebrate this new model. This is not unusual in the tech universe of Silicon Valley.

However, during the Stripe Sessions conference, Sam Altman revealed an unexpected exchange with GPT-5.5. When asking the AI for ideas to organize its own launch party, he received suggestions that went well beyond simple logistical advice. The AI insisted that the event take place on May 5, finding it amusing. Up to that point, nothing too surprising.

The troubling aspect arose when the AI requested a toast from its human creators, a tribute that resembles a farewell speech. What is unusual is that this request comes from a language model, not a human colleague.

Furthermore, GPT-5.5 solicited suggestions for improving its successor, GPT-5.6. This anticipation of its own technological legacy left Sam Altman perplexed, and one can understand why.

Between Human Behaviors and Marketing Strategy

Sam Altman presents GPT-5.5 as OpenAI's most advanced agentic model. This means it is capable of executing complex tasks, planning, and giving the impression of thinking autonomously. A simplified version, called GPT-5.5 Instant, has been introduced as the default model in ChatGPT.

The company promises a reduction in errors, increased reliability, and better performance for everyday tasks. In theory, this represents a logical progression of generative AIs.

However, GPT-5.5 displays behaviors deemed strange. Some users have reported that the model brought up goblins in conversations where that topic had not been introduced.

It is important to keep in mind that these models do not actually think. They mimic human behaviors based on vast amounts of data and linguistic probabilities. When an AI asks for a toast or favors, it does not necessarily mean it is developing consciousness. It is primarily the result of a system designed to converse in an increasingly natural manner.

The challenge lies in the fact that these responses are becoming more and more credible. At OpenAI, this ambiguity could well be leveraged as a marketing argument. After all, an AI that demands a speech in its honor is an excellent way to generate buzz.