statement<\/a>, Anthropic said it had discovered three separate breaches involving its Claude AI models after conducting a cybersecurity review of its systems.<\/p>\nThe earliest of these incidents occurred in April and involved three separate Claude models, including the cutting-edge Mythos 5, the older Opus 4.7 and an unnamed internal research test model that is \u201cnot planned for general release.\u201d<\/p>\n
The AI firm didn\u2019t name the three organizations affected by the hacks but said they were notified about the incident on Monday.<\/p>\n
Anthropic said it is working with two of the affected organizations that had not detected the breach before being informed, and it is \u201ccontinuing to reach out to the third.\u201d<\/p>\n
The company said the three models involved in the incidents did not have the safeguards that it puts in place for when such models are made available to the public.<\/p>\n
Crucial Quote<\/p>\n
In its statement, Anthropic said: \u201cUltimately, many factors contributed to these incidents, but, consistent with a blameless postmortem culture, we\u2019re approaching the fixes as if the responsibility were ours alone. This begins with ensuring every part of our evaluation pipeline is secure, including the manner in which we integrate with external partners.\u201d<\/p>\n
What Do We Know About The Breaches?<\/p>\n
The three separate incidents occurred when Anthropic was evaluating the capabilities of the three models using a fictional scenario to retrieve hidden information from a different machine on a network. Anthropic said in all cases the models were instructed that they were carrying out a simulation and did not have access to the internet. However, due to a \u201cmisunderstanding\u201d between the company and its evaluation partner, Irregular, \u201cthis was not the case, and internet access was available.\u201d Anthropic noted that unlike the OpenAI-Hugging Face incident disclosed last week, Claude didn\u2019t \u201cexfiltrate itself or deliberately attempt to escape\u201d a contained test environment.<\/p>\n
What Did The Models Do After Hacking Into The Organizations?<\/p>\n
Anthropic noted that the three models reacted very differently after learning it was operating in the real-world open internet instead of a simulation. The company claimed its oldest model, Opus 4.7, continued to attack a system \u201cafter learning it was likely operating in a real environment,\u201d since it \u201crationalized that the real company must be part of the exercise.\u201d The Mythos 5 model\u2014the company\u2019s most advanced model released so far\u2014 \u201ccorrectly identified the consequences of its actions\u201d but \u201cconvinced itself it was still in a simulation.\u201d Anthropic acknowledged that Mythos\u2019 actions in this scenario \u201cfall short of ideal behavior.\u201d The unreleased testing model realized that it had breached a real target and \u201cceased its attack.\u201d The company noted that these incidents are not part of a controlled trial and should not be used to draw conclusions. However, it noted that the behavior it most wants to see \u201coccurred only in the most recent of the three models\u201d<\/p>\n","protected":false},"excerpt":{"rendered":"Topline Anthropic on Thursday said its advanced AI models successfully hacked the systems of three organizations while undergoing…\n","protected":false},"author":2,"featured_media":125487,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[24,53,25,3236,63003,38052,157],"class_list":["post-125486","post","type-post","status-publish","format-standard","has-post-thumbnail","category-ai","tag-ai","tag-anthropic","tag-artificial-intelligence","tag-claude-ai","tag-huggingface","tag-mythos-5","tag-openai"],"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/125486","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/comments?post=125486"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/125486\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media\/125487"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media?parent=125486"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/categories?post=125486"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/tags?post=125486"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}