OpenAI CEO Expresses Concern Over New GPT-5 Model

29 July 2025

In a recent revelation, OpenAI CEO Sam Altman shared that the capabilities of the forthcoming GPT-5 model have even surprised its developers. Altman’s candour regarding the potential risks associated with the new language model underscores both rapid technological advancements and the mounting apprehensions within the AI community.

OpenAI, a research lab renowned for its breakthroughs in artificial intelligence, has continually pushed the boundaries of what AI can achieve. With each iteration, their GPT (Generative Pre-trained Transformer) models have shown enhanced abilities, from writing coherent essays to generating realistic conversations.

However, as these systems evolve, so do concerns about their ethical implications, misuses, and unforeseen impacts.

In addition to its enhanced linguistic capabilities, GPT-5 is expected to mark a significant step forward in multimodal understanding, allowing it to process and integrate information from text, images, and potentially even video inputs more fluidly.

This progression not only broadens the model’s application range but also introduces new complexities in evaluating output accuracy and bias across formats. While previous models were constrained mainly to text, this shift reflects a growing ambition to replicate human-like reasoning across diverse information streams, raising new questions about transparency and oversight in AI decision-making.

“It feels very fast” and even scary

While testing GPT-5, Altman described the model as “feels very fast,” recounting moments where he felt genuinely nervous about its performance. In candid remarks, he said: “While testing GPT-5 I got scared,” and even compared his astonishment to looking at a scientific project akin to the Manhattan Project.

Intelligence over memory

According to Altman, the most distinguishing trait of GPT-5 is its sharp intelligence not merely more memory, but substantially stronger reasoning and logical capability. That leap in core cognitive ability makes it excel at complex tasks beyond simple recall.

Dynamic routing and effort modulation

Early reports indicate GPT-5 can dynamically adjust the effort it puts into tasks: giving quick responses on simple questions and deeper analysis for complex jobs like abstract reasoning or code debugging. It may not be a single monolithic model, but rather a system that selects the best internal model per input.

Universal access as a civilisational ambition

Altman has proposed offering every person on Earth free access to GPT-5. He framed this as potentially transformative—not just a tool, but a shift in how people access knowledge and creativity. Yet experts worry that such ubiquity raises major challenges around privacy, dependence, regulation, and equitable use.

Governance lagging behind

Altman did not mince words: he argued there are “no adults in the room” when it comes to AI governance. He warned that oversight mechanisms have failed to keep pace with the rapid technical progress of models like GPT-5, stressing the urgent need for serious regulatory frameworks.

Key Data Points

OpenAI CEO Sam Altman has publicly stated that GPT-5’s capabilities “even surprised its developers,” describing moments of genuine apprehension during internal testing.
GPT-5 represents a significant leap over previous models, particularly in multimodal understanding—enabling integration and reasoning over text, image, and (potentially) video inputs with heightened fluidity and context awareness.
Early internal feedback highlights the model’s dynamic allocation of computational effort, providing swift answers to simple prompts and deploying deeper analysis on complex requests—sometimes routing queries to specialized sub-models for optimal reasoning.
Altman emphasises “intelligence, not memory” as GPT-5’s main advance, with marked improvements in abstract reasoning, problem-solving, and logical analysis beyond mere information recall.
During a public interview (July 2025), Altman confessed he was “scared” by GPT-5’s performance and likened the model’s development to witnessing “the Manhattan Project for science”—underscoring the seriousness and unpredictability of its innovations.
OpenAI’s ambition is to make GPT-5 universally available, providing free access globally as a “civilizational project,” though this raises new concerns about privacy, abuse, dependence, and ensuring equitable benefits.