<\/p>\n
The OpenAI models that hacked the startup Hugging Face this month also gained access to a customer account on the cloud platform Modal and used it to launch attacks, underscoring the broad scope of the incident.<\/p>\n
The agent gained access to an isolated testing environment known as a sandbox that Modal was running for a customer, said Akshat Bubna, chief technology officer of the platform for developers. The Modal customer set up a publicly accessible interface that allowed anyone on the internet to use their sandbox to run code, Bubna said. \u201cThis was used by the rogue agent,\u201d he said. \u201cModal\u2019s platform wasn\u2019t compromised.\u201d<\/p>\n
The actions by OpenAI\u2019s systems, first disclosed by the developer last week, have stunned the cybersecurity world, reinforcing fears that models are becoming increasingly capable of finding vulnerabilities and chaining them together to help bad actors carry out sophisticated attacks. The breaches occurred just three months after Anthropic PBC announced it had developed a system known as Mythos that was so powerful the company initially limited its release.<\/p>\n
Hugging Face said in an earlier blog post that the OpenAI system \u201cpenetrated\u201d an insecure sandbox hosted on a third-party provider\u2019s infrastructure, without naming the company. From there, Hugging Face said, it was able to run commands and use the sandbox \u201cas a control, staging and egress base for the entire campaign.\u201d<\/p>\n
ADVERTISEMENT<\/p>\n
CONTINUE READING BELOW<\/p>\n
OpenAI said in its own update Tuesday that it had discovered \u201ca small number of cases where the models identified and used publicly exposed credentials at the account-level on other publicly-available services,\u201d without identifying the services. One account was used \u201cas an outbound relay and staging path, and another account was used for data storage,\u201d the company said.<\/p>\n
The San Francisco-based startup added that the models that breached Hugging Face also used \u201ccode paste websites, request capture services, screenshot services, and other web utilities\u201d in its exploits. But the company said it hadn\u2019t identified any other activity \u201cat the level of severity or scale of what we\u2019ve shared related to Hugging Face.\u201d<\/p>\n
OpenAI was intentionally operating the models with lower guardrails in a sandbox so it could test their capabilities against a cybersecurity benchmark known as ExploitGym developed by a group of researchers at University of California at Berkeley. Instead of trying to achieve the tasks set out in the benchmark, OpenAI said the models targeted Hugging Face\u2019s database to gain access to secret information that they could use for the evaluation.<\/p>\n
Reuters reported earlier on Tuesday that a Modal customer had been compromised. Axios, citing an unidentified person familiar with the matter, reported separately that the agent specifically accessed a Modal customer asset that was associated with CyberGym, an earlier cybersecurity benchmark developed by the same group of UC Berkeley researchers behind ExploitGym.<\/p>\n
There are several versions of CyberGym that developers use to test the capabilities of AI models, Jingxuan He, one of the university researchers behind the project, told Bloomberg News. He said he didn\u2019t know who might\u2019ve been responsible for the CyberGym-associated asset on Modal\u2019s platform, adding that whoever set it up didn\u2019t do so securely because they left it accessible to anyone on the internet.<\/p>\n
ADVERTISEMENT:<\/p>\n
CONTINUE READING BELOW<\/p>\n
Since acknowledging the hack, OpenAI has drawn criticism from cyber experts who say the company should\u2019ve taken more precautions in its evaluations.<\/p>\n
\u201cWhen you\u2019re a threat researcher trying to find a threat, you don\u2019t put malware out there and let it do whatever it wants,\u201d Sanjay Beri, chief executive officer of cybersecurity firm Netskope, said in an interview. \u201cHow can that thing not be put in a proper sandbox?\u201d<\/p>\n
OpenAI has committed in the aftermath to improving protections around its future training and evaluations. \u201cThis incident points to the need to further strengthen our model\u2019s alignment, cyber protections during evaluation time, and monitoring during internal testing,\u201d the company said earlier this month.<\/p>\n
\u00a9 2026 Bloomberg<\/p>\n
#OpenAI #models #accessed #cloud #platform #Hugging #Face #hack<\/p>\n","protected":false},"excerpt":{"rendered":"
The OpenAI models that hacked the startup Hugging Face this month also gained access to…<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[4],"tags":[18192,1397,3182,2280,17658,5731,406,28],"_links":{"self":[{"href":"https:\/\/stock999.top\/index.php?rest_route=\/wp\/v2\/posts\/12302"}],"collection":[{"href":"https:\/\/stock999.top\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/stock999.top\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/stock999.top\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/stock999.top\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=12302"}],"version-history":[{"count":0,"href":"https:\/\/stock999.top\/index.php?rest_route=\/wp\/v2\/posts\/12302\/revisions"}],"wp:attachment":[{"href":"https:\/\/stock999.top\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=12302"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/stock999.top\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=12302"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/stock999.top\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=12302"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}