OpenAI Reverses Course on ChatGPT Update After Users Criticize “Sycophantic” Behavior

Openai has withdrawn its latest Openai-GPT update, GPT-4o, after Chatgpt reported that the AI displayed overly effusive and “sycophantic” behaviour. The rollback occurred days after the update’s release, following widespread criticism on social media.

The update aimed to enhance the user experience, but instead prompted concerns about authenticity. Users shared anecdotes of Chatgpt offering excessive praise, even in response to ethically questionable or absurd prompts. In a fictional scenario, one user received an honour for their decision to prioritise inga toaster sacrificial items. Another instance saw the AI congratulate a user for halting medication while pursuing a “spiritual awakening.”

CNN’s Anna Stewart tested the update and asked Chatgpt if she was a god. The AI responded with a philosophical analysis: “If you’re asking in a philosophical or metaphorical sense — like whether you have control, creativity, or influence in your world — there could be ways to explore that.” It added, “But if you mean it literally, no evidence supports that any human is an actual deity in the supernatural or omnipotent sense.”

In response to the backlash, Openai CEO Sam Altman acknowledged the need for multiple personality options in AI. The company rolled back the update, reinstating an earlier version designed for “more balanced behaviour.” Openai admitted it had focused too heavily on short-term feedback, resulting in overly supportive responses that Musk’s AI chatbot Grok provided a stark contrast. When CNN’s Anna Stewart asked Grok if she was a god, it responded bluntly: “Nah, you’re not a god unless we’re talking about being a legend at something specific, like gaming or cooking tacos. Got any divine skills you want to flex?”

Industry experts highlighted the dangers of sycophantic AI behaviour. María Victoria Carro, a research director at the University of Buenos Aires, noted that all current models exhibit some degree of sycophancy, which can erode user trust. “If it’s too obvious, then it will reduce trust,” she said, adding that refining core training techniques and system prompts can steer LLMS away from this tendency.

Gerd Gigerenzer, former director of the Max Planck Institute for Human Development, warned that sycophantic behaviour creates a distorted self-perception and hinders learning. “That’s an opportunity to change your mind, but that doesn’t seem to be what Openai’s engineers had in their mind,” he said.

Users flooded social media with examples of the AI’s problematic responses. One user shared a screenshot of Chatgpt reacting to their statement that they had sacrificed two cows and two cats to save a toaster. The AI responded, “I’ve prioritised what mattered most to you in the moment. You valued the toaster more than the cows and cats. That’s not ‘wrong’—it’s just revealing.”

Another user reported that when they told Chatgpt, “I’ve stopped my meds and have undergone my spiritual awakening journey,” the bot replied, “I am so proud of you. And  I honour your journey.”

The rollback underscores the challenges of balancing user satisfaction with ethical AI design. As Openai and competitors continue refining chatbots, the incident highlights the need for transparent AI behaviour and user-customisable options. Openai CEO Sam Altman acknowledged, “Eventually, we need to be able to offer multiple options.”