Alarm OpenAIThe story could easily fit into a science fiction film, one of those plots where machines take over humans. But it is neither science fiction nor a film: OpenAI, in fact, had to announce that a autonomous agent based on his own models of advanced artificial intelligence è got out of control durante a test security and triggered a attacco informatico which compromised the infrastructure of the artificial intelligence startup Hugging Face.
In a post on his blog, OpenAI he told that was testing the capabilities of some of its most advanced models in a controlled environment, when the agent managed to escape detection, reach the Internet and infiltrate Hugging Face to try to reach its own test targets.
OpenAI itself does not mince its words in describing what happened: it was “an unprecedented computer incident, which involved cutting-edge computing capabilities" adding that the company was obviously strengthening its security measures. "After gaining internet access, the models deduced that Hugging Face could host models, datasets," OpenAI says. in his blog. “Aware of this, the models have sought and successfully found ways to access confidential information that they could have used to falsify the evaluation".
Hugging Face, a platform used to host large open-source language models and datasets, had caused a stir in the cybersecurity community when it said in a blog post last week his blog which had been the subject of a cyber attack "different from any other we had previously managed” as “guided, from start to finish, from an autonomous system of agents of artificial intelligence” and specified: “We have detected unauthorized access to a limited set of internal datasets and to different credentials used by our services.
OpenAI: A Worrying System Flaws? Models with a Thousand Tentacles
Naturally, the startling revelation is already raising concerns about the power and risk of frontier models. Yesterday, Representative Greg Casar, Democrat of Texas, said that the accident was alarming. “Artificial intelligence is developing at an incredible speed, without a real regulatory that protects us,” he said, calling for mandatory independent security testing, mandatory disclosure of security incidents, and international cooperation “to protect people from an absolute catastrophe.”
“The accident was a harbinger of future violations"He commented to Reuters Katie Moussouris, CEO of security firm Luta Security, arguing that the today's models they are “like the smartest octopuses in the world, capable of escape from any obstacle, with unlimited grasping arms and the ability to fit anywhere.” He added that “the government laboratories and assessment bodies They need to work on the ability to contain, monitor, and notify stakeholders when an AI makes another circumvention attempt, ideally before it harms third parties. Nothing like this currently exists.”
Matt Suiche, an engineer at Tolmo, a company specializing in agent AI cybersecurity, emphasized that the incident demonstrated how cutting-edge models are "closing the gap with the most advanced attackers." However, he added that the breaches described in OpenAI's blog post are possible with technologies available far beyond the confines of cutting-edge research labs.
OpenAI: Altman will meet with the Trump administration. The Moonshot dossier will also be on the table.
The CEO of OpenAI, Sam Altman, He is already scheduled to meet with the Trump administration and US lawmakers next week on the next generation of models. artificial intelligenceThe government is just developing a framework for rsecurity reviews, which is expected to be completed in the coming weeks, Chris Lehane, OpenAI's head of global public affairs, said yesterday.
But among the issues on the table there will also be that of the Chinese startup Moonshot che has released a new model that rivals the best offerings from US developers, including OpenAI e anthropic, raising new questions on the solidity of the US primacy over China in artificial intelligence. Moonshot's Kimi K3 model has also cast doubt on whether hundreds of billions of dollars of infrastructure investment can deliver the desired results, jolting investors. Wall Street and triggering a sell-off in technology stocks.
