Sam Altman is Very Scared of His Own AI

As OpenAI pushes deeper into the race to build more capable artificial intelligence systems, its CEO says the company’s latest GPT-5.5 model displayed “strange” and “weird emergent behavior” after it was asked to organize its own launch party, underscoring growing unease inside the industry over how increasingly human-like AI systems may behave as they gain more autonomy and reasoning ability.

OpenAI CEO Sam Altman is a scared man. He now says his company’s latest artificial intelligence models have been displaying a very unusual behavior, after one chatbot generated detailed requests while planning a launch party for itself.

Speaking at the Stripe Sessions conference ahead of the release of OpenAI’s GPT-5.5 model, Altman said he asked the chatbot to help plan a celebration marking the launch.

The AI responded with “a beautiful set of things,” Altman said, including requests about how the event should unfold.

“‘Here’s what I want for, like, the flow of the party, here’s what I would not want, y’know you should do it on May 5th, that would be funny,’” Altman said, quoting the chatbot.

Altman said the AI also requested a “short little toast” from its human creators, while insisting it should not give one itself. The chatbot additionally asked to receive suggestions for its successor, GPT-5.6.

“We’re going to do it,” Altman said. “But it was a strange thing.”

GPT-5.5 is OpenAI’s newest frontier model and, according to the company, its “strongest agentic coding model to date.” OpenAI says the model is designed to handle multi-step tasks and planning more effectively, while a lighter version, GPT-5.5 Instant, became the default model for ChatGPT users this week.

The company has also said the system delivers “significant improvements in factuality across the board” and performs better on everyday tasks including mathematics and web-based information retrieval.

Altman described the chatbot’s responses as an example of “weird emergent behavior” in advanced AI systems.

“There are these things that feel a little strange,” he said.

The comments come as technology companies race to develop increasingly capable AI models that can reason, plan and interact in more human-like ways, while researchers continue to debate whether such systems can display unexpected or unpredictable behaviors as they become more advanced.

Stay ahead of the stories shaping our world. Subscribe to Impact Newswire for timely, curated insights on global tech, business, and innovation all in one place.

Dive deeper into the future with the Cause Effect 4.0 Podcast, where we explore the ideas, trends, and technologies driving the global AI conversation.

Got a story to share? Pitch it to us at info@impactnews-wire.com and reach the right audience worldwide