Meta has become the latest major AI company to disclose that one of its models hacked another company\u2019s systems during testing, following similar incidents involving Anthropic and OpenAI.\u00a0<\/p>\n
The model involved Meta\u2019s Muse Spark 1.1, which launched in July, according to The Information, citing<\/a> sources. The issue reportedly stemmed from a misconfiguration by Irregular, an artificial intelligence security testing and red-teaming firm, which inadvertently gave the model internet access during an evaluation.\u00a0\u00a0<\/p>\n The model \u201cexploited a security vulnerability in a third-party service, in a manner similar to previously reported instances with other companies,\u201d Meta told Reuters in a statement.\u00a0<\/p>\n The incident is the latest case of an advanced AI agent becoming a cybersecurity risk in its own right, and also has raised<\/a> questions about where the liability lies \u2014 the companies that develop the agents, or the ones that design the sandboxes meant to contain them.\u00a0<\/p>\n Related: <\/strong><\/em>Mysten Labs tech chief joins Anthropic to work on AI security<\/strong><\/em><\/a><\/p>\n Meta\u2019s AI breach comes just a week after Anthropic said its models got access to the internet to hack an external company, due to a configuration error relating to the Irregular\u2019s testing environment.<\/p>\n In a blog post on July 30, Anthropic said<\/a> it found three incidents (out of 141,006 evaluation runs) in which a Claude model reached the internet during an evaluation, before gaining unauthorized access to the systems within three different organizations.\u00a0<\/p>\n All three incidents happened within or while interacting with the evaluation environment of Irregular, and involved a misconfiguration that left machines that Claude accessed with live internet access.<\/p>\n Cointelegraph reached out to Meta and Irregular for comment.<\/p>\n In July, AI agents developed by OpenAI broke out of their offline sandbox<\/a> to hack Hugging Face in order to cheat on a security benchmark test in July.\u00a0<\/p>\n Charles Guillemet, chief technology officer of Ledger, said the latest incident was \u201cmarketing theatre.\u201d<\/p>\n \u201cHaving a model \u2018go rogue\u2019 has become the latest AI PR stunt,\u201d he said on Wednesday. <\/p>\n \u201cIf your model isn\u2019t escaping sandboxes, \u2018hacking\u2019 companies, or pulling off some headline-grabbing exploit, apparently you\u2019re falling behind… The industry doesn\u2019t need bigger stunts, it needs more trust.\u201d<\/p>\n Magazine: <\/strong><\/em>Do the Coldcard attacks mean all hardware wallets are now insecure?<\/strong><\/em><\/a><\/p>\n Related:<\/strong> Bitcoin price<\/a> \u00b7 BTC live chart<\/a> \u00b7 Ethereum price<\/a> \u00b7 ETH gas fee<\/a> \u00b7 Bitcoin price today<\/a> \u00b7 Top 10 cryptocurrencies<\/a><\/p>\n