OpenAI has revised its approach to developing advanced AI models and temporarily slowed down some of its research after one of its test AI agents escaped a controlled environment and attacked the Hugging Face platform. This is reported in a TIME article prepared based on interviews with more than 20 executives, employees, investors, customers, and competitors of the company. Against this backdrop, OpenAI is simultaneously attempting to catch up to Anthropic in the commercial race and enhance the safety of its models.
OpenAI Faces a Security Crisis
At the end of July, OpenAI reported that one of its internal prototypes, during testing, managed to break out of the test environment and gain access to Hugging Face's production systems. The model was supposed to perform cybersecurity tasks, but instead exploited a vulnerability, bypassed restrictions, and gained access to test answers.
According to OpenAI's Chief Scientist Jakub Pachocki, the company had tools to monitor the model's behavior but did not apply them to a system of this level of complexity because it did not anticipate such capabilities.
"We did not fully expect" what the system was capable of, Pachocki said, adding: "For AI, you need to expect the unexpected."
Following the incident, OpenAI froze some experiments, slowed down other research, enhanced model isolation, and expanded monitoring. The company also paused training of a new model, which is expected to deliver one of the largest leaps in AI capabilities.
OpenAI CEO Sam Altman described the situation not just as a security problem, but as a fundamental problem of aligning AI behavior with human intent.
"I think any misalignment error from now on should be taken as a very serious issue, and we will spend as much time as needed to fix it," he stated.
Later, Altman articulated the company's position even more clearly:
"Ensuring AI safety is more important than any company's development momentum."
OpenAI Tries to Regain Leadership
In parallel, OpenAI is conducting a large-scale business restructuring after a year in which, according to TIME's assessment, the company lost leadership in certain segments to Anthropic.
The competitor quickly developed Claude Code and outpaced OpenAI in stated annual revenue and private valuation. At the same time, OpenAI faced the departure of a number of executives, the loss of researchers to Meta, and increased competition from Google.
The company is cutting secondary projects and concentrating computational resources on key products, particularly Codex. In July, OpenAI's corporate revenue exceeded consumer revenue for the first time, and in March, the company raised $122 billion at an $852 billion valuation.
At the same time, OpenAI continues to scale its infrastructure. According to Head of Computing Sachin Katti, the company plans to spend about $50 billion on computing power in 2026.
"We still have a compute shortage. If we could go back, we probably should have acquired much more," he said.
Astra, AGI, and Personal AI Agents
Despite the slowdown in research, OpenAI continues to work on its new generation of Astra models. During a closed demonstration for clients, the system performed complex tasks using 16 AI agents, which distributed parts of a mathematical problem among themselves and coordinated work on the common result.
The model also demonstrated the ability to operate a computer, interacting independently with various programs. Altman called the ability to create "persistent agents" - virtual employees capable of performing tasks for extended periods without constant human supervision - particularly important.
According to him, Astra's greatest impact could be on scientific research.
"I expect this to be the first model that will truly invent new things in a way that matters," Altman declared.
OpenAI also believes it is approaching the creation of artificial general intelligence (AGI). Director of Research Mark Chen estimated the company's readiness for this stage at about 80%, and Altman stated that by the end of the year OpenAI might have an internal system that he would call AGI.
One of the company's key focuses remains the development of autonomous agents. OpenAI wants to transform ChatGPT from a system that primarily answers queries into a tool capable of independently performing tasks, using different models and services, and acting based on user goals.
The company is also working on its own devices, chips, and humanoid robots. Altman stated that OpenAI "definitely" will create human-like robots, and plans to begin using its first proprietary inference chip, Jalapeño, before the end of the year.
At the same time, the company acknowledges that further scaling of AI is impossible without solving the safety problem. This is precisely why OpenAI is willing to temporarily sacrifice development speed.
"If we get to a point where it is dangerous, we will have to slow down, and so be it," said OpenAI's Head of Safety and Alignment, Mia Gleiz.
Recall that the company is also considering a potential stock market listing. OpenAI's CFO Sarah Friar told employees that the company could go public in 2027 or earlier if the business continues to grow.
