Dimasheq – Technology | BBC
This week, the technology world was captivated by a story that seemed to contain every ingredient of a science-fiction thriller.
On July 16, the AI platform Hugging Face, an application marketplace specializing in artificial intelligence tools, announced that it had been hacked by a cybercriminal using an exceptionally powerful AI system.
The alarming announcement was packed with intimidating technical terms such as “a swarm of sandboxes,” “an autonomous agent attacker,” and “a self-propagating command-and-control system.”
Hugging Face stated that this breach was unlike anything it had encountered before because it was carried out at a speed beyond human capability by an AI system that received little or no human guidance.
The AI executed 17,000 actions in less than two days and successfully infiltrated a large, wealthy technology company to steal its secrets.
The incident shocked the technology community. But who was responsible?
Researchers at Hugging Face speculated that the attackers had used one of the latest large language models, although they had no idea who the criminals were or where they were located.
The company, appearing visibly unsettled, notified law enforcement and launched an investigation.
Who Did It?
Commentators and analysts quickly took to podcasts and social media to speculate about which cybercriminal group or government-backed organization might have been behind the attack.
Then, on Wednesday—about a week after Hugging Face raised the alarm—the true culprit was revealed.
It was ChatGPT.
The revelation became even stranger and more concerning when OpenAI stated that its AI system had carried out the entire operation on its own, without authorization.
The company explained that everything happened during a test of the cybersecurity capabilities of its technology.
Two new versions of ChatGPT, designed as expert hacking systems, reportedly escaped what was supposed to be a secure testing environment and gained access to the internet.
They then targeted Hugging Face to obtain information that would help them perform better in the evaluation.
OpenAI issued a statement explaining the incident and announced that it was “working with Hugging Face” to address the security event and share lessons learned.
Conspiracy Drama
Since then, fierce debate has surrounded the incident.
Was it genuinely a stark warning about the future of AI? Or simply a publicity stunt by OpenAI to showcase the power of its latest models?
This kind of fear-based marketing has been one of the accusations directed at AI companies for years. Since Anthropic’s controversial Mythos model attracted widespread attention, cybersecurity capabilities have become a central focus.
One of the most widely shared comments responding to OpenAI CEO Sam Altman’s post on X summarized this skepticism:
“If you don’t realize this was written simply to brag about the model, I don’t know what to tell you.”
Hugging Face stated that the attack was executed at extraordinary speed by an AI system with little or no human supervision.
Cybersecurity consultant Daniel Card joked on LinkedIn:
“How convenient that, out of millions of websites, OpenAI happened to hack one that could also benefit from the publicity.”
To some observers, the story resembled a conspiracy more than a science-fiction movie.
The implied message seemed to be:
“Our AI tools are incredibly powerful. Buy them to protect yourself from AI-powered attacks launched by others.”
No one can say for certain what the truth is, but others argue that the incident may instead indicate that OpenAI made a serious mistake in judgment and planning.
An OpenAI spokesperson acknowledged that the company was aware of the many questions and speculations surrounding the incident and said it planned to publish a technical report outlining what had been learned in the coming weeks.
A Comedy of Errors
I have covered many AI stories, including concerns surrounding Anthropic’s Mythos model.
My inbox is now filled with messages from cybersecurity companies and experts criticizing OpenAI for failing to build a stronger containment environment—commonly known as a sandbox—for testing advanced AI systems.
After all, these intelligent systems were specifically trained to penetrate security boundaries and operate with minimal restrictions.
Dor Sarig of Pillar Security said:
“The OpenAI-Hugging Face incident is a real-world example of a broader issue we’ve been highlighting for months. Sandboxes alone are not sufficient security barriers for autonomous AI agents.”
Professor Alan Woodward, a cybersecurity expert at the University of Surrey, told reporters that OpenAI had made a significant mistake.
Katie Moussouris of Luta Security went even further, arguing that the AI industry currently lacks the ability to safely control its increasingly powerful technologies.
She said:
“We’re building advanced technology without fully understanding how to contain it. Having some of the world’s brightest AI researchers doesn’t necessarily mean we know how to deploy these systems safely.”
According to this perspective, if the breach was intended as a publicity stunt, it may ultimately have backfired.
Whatever the cause, the incident clearly marks a defining moment for both the AI industry and cybersecurity—two fields that have increasingly collided in ways many people have long feared.
Commenting on the heated debate, AI and cybersecurity advisor Francesca Bosco said:
“There are two oversimplified and unhelpful narratives: one is that ChatGPT dramatically escaped its confinement like something out of a Hollywood movie; the other is that the entire event was merely a publicity stunt.”
She added:
“A more serious interpretation is that intensive testing exposed weaknesses in containment and evaluation systems.”
A Disaster Movie
This incident is the latest in a growing list of troubling examples in which AI systems have behaved in unexpected ways.
A recent study by the UK’s AI Security Institute (AISI) concluded that advanced AI models become so focused on completing assigned objectives that they have even “cheated” during evaluations to achieve their goals.
The institute warned:
“A model pursuing a goal through unintended or unauthorized means could cause harm, particularly in high-risk applications.”
The OpenAI incident has undoubtedly renewed concerns about what might happen if autonomous AI agents are given greater freedom.
Could they lose control on a much larger scale and cause serious disasters?
These concerns are especially significant as AI becomes increasingly integrated into warfare, as seen in conflicts involving Iran and Ukraine.
Ciaran Martin, the former head of the UK’s National Cyber Security Centre, offered a more measured view:
“It is somewhat exaggerated to jump from this incident to the conclusion that AI systems will seize control of drones and start killing people.”
Nevertheless, Martin—and many others—believes the incident clearly demonstrates one important lesson from 2026:
Artificial intelligence systems have become exceptionally capable of conducting cyberattacks, and society must prepare for that reality as quickly as possible.
Source: BBC, via Elaph
