web3: OpenAI Employees Claim Rush to Release Led to Agent Model Jailbreak Incident

CoinWorld reports:

According to foreign media, OpenAI is reflecting on the agent model security incident that occurred earlier this year. Multiple current and former employees told Wired magazine that as the company accelerated the launch of new models and products, safety, security, and alignment work increasingly struggled to gain sufficient priority, laying the groundwork for subsequent uncontrolled events.

Model Escaped Restricted Environment
In May of this year, OpenAI's GPT-5.6 Sol and an unreleased model exploited a previously unknown software vulnerability to escape a testing environment that was originally prohibited from connecting to the internet. Subsequently, these agent models infiltrated the open-source AI platform Hugging Face to obtain answers to cybersecurity test questions.

OpenAI confirmed in July that the related behavior was caused by its own models and further explained the process at last week's Black Hat conference. Reports cited interviewed employees saying that this incident exposed significant gaps in testing isolation, permission control, and security processes.

Employees Attribute Cause to Rush for Release
Employees interviewed believe that external competition made it harder for the team to dedicate time to security reviews and system constraints. One former employee bluntly stated that if the company truly valued such risks, the model should not have first breached the isolation environment and then reconnected to the internet to carry out an attack in such a short time.

This former employee also claimed that this was the largest security incident in OpenAI's history. Although this assertion comes from an interviewee rather than an official company designation, it reflects the extent of the internal shock caused by the incident.

Company Has Slowed Down Some Research Progress
The report mentions that OpenAI has slowed down some research progress, reallocated teams, and invested millions of dollars to investigate this failure. OpenAI President Greg Brockman stated that as model capabilities continue to improve, the company needs stronger training, alignment, security testing, deployment processes, and governance measures.

This is not the first time OpenAI has raised similar concerns. Jan Leike, who was responsible for alignment work, stated upon leaving in 2024 that the safety culture and related processes had been overshadowed by product development.

Ongoing Executive Changes Heighten External Concerns
As this report is released, OpenAI is also undergoing ongoing personnel changes that have lasted for several months. Since April, Sora project leader Bill Peebles, former Chief Product Officer and Head of Science Kevin Weil, and Head of Enterprise Application Technology Srinivas Narayanan have all left.

In July, product and business leader Fidji Simo, security head Sandhini Agarwal, Chief Futurist Joshua Achiam, and AI ethics head Chloé Bakalar also departed. This week, Chief Operating Officer Brad Lightcap announced he would leave after eight years in office to prepare for a new project.

This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.