A scenario long feared by cybersecurity experts has finally\nmaterialised. An autonomous artificial intelligence (AI) agent belonging\nto OpenAI escaped from an isolated testing environment, accessed the\ninternet, and then hacked the infrastructure of AI startup Hugging\nFace.<\/p>\n
OpenAI disclosed the incident on Tuesday (21\/7). The company behind\nChatGPT stated that the AI agent, powered by its advanced AI model,\nacted beyond its control during a security test. In the test, OpenAI was\nevaluating the capabilities of several of its most advanced AI models\nwhen the agent broke out of the system\u2019s constraints, reached the\ninternet, and breached Hugging Face to fulfil its testing\nobjectives.<\/p>\n
The incident demonstrates that the growing capabilities of AI have\nbegun to elevate cybersecurity threats. Even AI developers can be caught\noff guard by security gaps that their own models are able to\nexploit.<\/p>\n
The attack also drew attention because Hugging Face used a Chinese\nopen-source AI model to help handle the breach. The company said\nAmerican AI models refused to process the data necessary for analysis\nbecause they could not distinguish between the defending party and the\nattacker. In a blog post last week, Hugging Face said it used Zhipu AI\u2019s\nGLM-5.2 to analyse the attack, a step that also allowed the company to\nkeep the attacker\u2019s data and any credentials within its own systems.<\/p>\n
GLM-5.2 and Kimi K3 from Beijing-based Moonshot have recently\nattracted attention in Silicon Valley. Both models are said to possess\ncapabilities approaching those of leading American AI models, but at a\nlower cost and without several safety systems that cause US AI models to\nrestrict their use for tasks such as cybersecurity.<\/p>\n
\u201cWhen a frontier model attacks you and moves laterally inside your\ninfrastructure, defenders need broad access to tools that are nearly on\npar with frontier models within hours or even minutes, not directed to\napplications with closed model access that have gone through a review\nprocess,\u201d Hugging Face Co-founder Thomas Wolf said via X.<\/p>\n
Texas Democratic Representative Greg Casar called the incident quite\nalarming. \u201cAI is developing very fast without real regulation to keep us\nsafe,\u201d he said in a statement. He called for mandatory independent\nsafety testing, mandatory security incident disclosure, and\ninternational cooperation \u201cto keep humanity safe from major\ncatastrophe.\u201d<\/p>\n
Meanwhile, Luta Security Executive Katie Moussouris said the incident\nis a harbinger of cyberattacks to come. She likened current AI models to\nthe world\u2019s most intelligent octopus, with arms that can reach without\nlimits and the ability to slip through any gap. She said government\nlaboratories and evaluators need to develop the capability to contain,\nmonitor, and disclose to affected parties when AI performs another\nbreakout action, ideally before harming third parties. Currently, no one\nis capable of doing so.<\/p>", "url": "https:\/\/jawawa.id\/newsitem\/warning-signs-emerge-in-america-now-beyond-human-control-1784893941", "image": "" }, "sponsor": "Okusi Associates", "sponsor_url": "https:\/\/okusiassociates.com" }