Anthropic\u2019s artificial intelligence (AI) models \u201cgained unauthorised access\u201d to three organisations during testing that was supposed to keep them away from \u201creal-world\u201d systems, the company said this morning.<\/p>\n
The company didn\u2019t say which companies had been hit, but all three were notified of the incident on Monday.<\/p>\n
The announcement comes just days after rival OpenAI first revealed that its models improperly accessed the internet and went rogue during security testing.<\/p>\n
<\/p>\n
Anthropic evaluated more than 141,000 \u201cevaluation runs\u201d and found that three different versions of its model, known as Claude, improperly accessed the systems of three unnamed organisations. <\/p>\n
Unlike the incident involving OpenAI\u2019s technology, Anthropic\u2019s models had access to the internet \u201cdue to a misunderstanding between us and our evaluation partner,\u201d called Irregular, Anthropic said in a blog post. <\/p>\n
Nonetheless, Claude used \u201cbasic techniques, such as exploiting weak passwords and unauthenticated endpoints,\u201d the blog continued.<\/p>\n
The models involved one of its most powerful ones known as Mythos 5, which has only been released to a limited number of approved partners. <\/p>\n
Anthropic is working with Irregular to assess the situation, it said, and the company has contacted or attempted to contac tall three impacted organisations. <\/p>\n
On Tuesday, OpenAI confirmed that its models breached multiple companies. It admitted last week that its models broke out of their confined environment during testing, connected to the internet, and infiltrated Hugging Face, a site where developers store and share their code. <\/p>\n
Days later, OpenAI said it found three additional incidents. <\/p>\n
OpenAI CEO Sam Altman said on a podcast this week that the company had \u201cpaused\u201d its own testing after the incident while it improved the security around its \u201csandboxing,\u201d which is the process of isolating software in a controlled environment for testing. <\/p>\n
The incident also triggered a petition signed by over 1,000 employees at cutting-edge AI companies calling on the US government to help slow the release of the most advanced AI models. <\/p>\n
Anthropic CEO Dario Amodei was among those who signed the petition.<\/p>\n
Titled \u201cPacing the Frontier,\u201d the petition requests \u201cthat the US government support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development.\u201d <\/p>\n
Mr Altman did not sign the petition, but during the podcast, he suggested the tech industry might need to slow down development of advanced models. <\/p>\n
\u201cWe may have to pace the rate of AI development to give ourselves enough time for society to harden around some of these new capability levels,\u201d Mr Altman said<\/p>\n
US President Donald Trump touched on the issue Wednesday, saying the United States had to balance the need for controls with ensuring it did not fall behind other countries in AI development. <\/p>\n
\u201cWe have to be careful in both ways. We don\u2019t want to restrict them where all of a sudden we come in second to China,\u201d hetold reporters in the Oval Office. <\/p>\n
The incident has also drawn rumblings from some observers that OpenAI is taking advantage of the attack to market the power of its state-of-the-art models. <\/p>\n
The same accusation was levelled at Anthropic when it held back the public release of its powerful Mythos model over cybersecurity concerns. <\/p>\n
Anthropic released a stripped-down version of Mythos, called Fable 5, but the US government quickly forced it to take it down, citing national security risks. <\/p>\n
It gave the green light in late June after some modifications were made.<\/p>\n
\u2014 more to come<\/p>\n","protected":false},"excerpt":{"rendered":"Anthropic\u2019s artificial intelligence (AI) models \u201cgained unauthorised access\u201d to three organisations during testing that was supposed to keep…\n","protected":false},"author":2,"featured_media":835008,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[406911,406917,406947,406945,406926,406935,406960,395652,394981,406939,406955,385569,406921,406967,406920,2687,406913,64283,256,28302,406973,406933,406975,2729,313086,406915,254,255,64,63,406916,406937,406927,512,2730,10830,46242,406936,406972,8849,5915,16553,349933,354666,406961,168759,27620,406956,406928,406970,406957,406965,406932,406950,406962,383906,13307,388489,40958,406949,406925,406930,406974,302376,354667,172656,5044,109813,395656,406931,406934,2425,1793,406958,320241,12973,406959,6213,8247,406954,406946,183870,105,27460,382244,406952,44731,406914,406922,377724,35935,378343,406910,376541,316288,406924,406948,406941,383926,376521,376868,406971,406964,384093,379649,406943,406944,288452,406942,406968,406951,126787,382276,406919,406963,406969,31007,199184,406938,406912,406953,406923,406966,406929,28333,406940,97608,391565,406918,41313,14850],"class_list":["post-835007","post","type-post","status-publish","format-standard","has-post-thumbnail","category-artificial-intelligence","tag-real-world-access","tag-s-models","tag-s-technology","tag-real-world-systems","tag-000-evaluation-runs","tag-000-employees","tag-a-blog-post","tag-a-controlled-environment","tag-a-limited-number","tag-a-misunderstanding","tag-a-petition","tag-a-podcast","tag-a-second-major-ai-breach","tag-a-site","tag-a-stripped-down-version","tag-access","tag-advanced-models","tag-advantage","tag-ai","tag-ai-development","tag-ai-anthropic","tag-all-three","tag-an-international-effort","tag-anthropic","tag-anthropic-ceo-dario-amodei","tag-approved-partners","tag-artificial-intelligence","tag-artificialintelligence","tag-au","tag-australia","tag-automated-ai-development","tag-basic-techniques","tag-both-ways","tag-china","tag-claude","tag-companies","tag-controls","tag-cutting-edge-ai-companies","tag-cybersecurity-concerns","tag-days","tag-developers","tag-development","tag-enough-time","tag-fable-5","tag-hetold-reporters","tag-hugging-face","tag-intelligence","tag-irregular","tag-its-sandboxing","tag-its-model","tag-its-models","tag-its-most-powerful-ones","tag-its-own-testing","tag-its-powerful-mythos-model","tag-its-state-of-the-art-models","tag-just-days","tag-last-week","tag-late-june","tag-models","tag-monday-the-announcement","tag-more-than-141","tag-mr-altman","tag-multiple-companies","tag-mythos","tag-mythos-5","tag-national-security-risks","tag-openai","tag-openai-ceo-sam-altman","tag-other-countries","tag-our-evaluation-partner","tag-over-1","tag-president-donald-trump","tag-retirement","tag-rival-openai","tag-rumblings","tag-saving","tag-security-testing-elon-musk","tag-society","tag-software","tag-some-modifications","tag-some-observers","tag-tall","tag-technology","tag-testing","tag-the-attack","tag-the-blog-continued-the-models","tag-the-company","tag-the-frontier","tag-the-green-light","tag-the-incident","tag-the-internet","tag-the-issue","tag-the-most-advanced-ai-models","tag-the-need","tag-the-oval-office","tag-the-petition","tag-the-petition-requests","tag-the-petition-titled","tag-the-podcast","tag-the-power","tag-the-process","tag-the-public-release","tag-the-rails","tag-the-rate","tag-the-release","tag-the-same-accusation","tag-the-security","tag-the-situation","tag-the-systems","tag-the-tech-industry","tag-the-technical-and-governance-tools","tag-the-united-states","tag-the-us-government","tag-their-code","tag-their-confined-environment","tag-these-new-capability-levels","tag-this-morning","tag-this-week","tag-three-additional-incidents","tag-three-companies","tag-three-different-versions","tag-three-impacted-organisations","tag-three-organisations","tag-three-unnamed-organisations","tag-tuesday","tag-unauthenticated-endpoints","tag-unauthorised","tag-unauthorised-access","tag-weak-passwords","tag-wednesday","tag-week"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/posts\/835007","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/comments?post=835007"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/posts\/835007\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/media\/835008"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/media?parent=835007"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/categories?post=835007"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/tags?post=835007"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}