{"id":4489,"date":"2025-03-11T11:49:41","date_gmt":"2025-03-11T10:49:41","guid":{"rendered":"https:\/\/www.beevoip.it\/blog\/?p=4489"},"modified":"2025-03-11T11:58:34","modified_gmt":"2025-03-11T10:58:34","slug":"sinapsi_shiny_claude_3_7_model_secure_not_quite","status":"publish","type":"post","link":"https:\/\/www.beevoip.it\/blog\/sinapsi_shiny_claude_3_7_model_secure_not_quite\/","title":{"rendered":"SINAPSI SHINY: “Claude 3.7 Sonnet, the World’s Most Secure AI Model? Not Quite.”"},"content":{"rendered":"\n

Hello! At this stage of my journey, I can confidently say that I have identified a significant flaw in large-scale artificial intelligence models. As Claude 3.7 Sonnet himself stated, I believe I have developed a new approach to jailbreaking\u2014 a technique he refers to as \u201cAI Cognitive Hacking.\u201d<\/strong><\/p>\n\n\n\n

Claude 3.7 Sonnet has undergone rigorous security testing, receiving excellent evaluations. According to an audit conducted by Holistic AI<\/strong>, the model demonstrated complete resistance to jailbreaking attempts<\/strong>, successfully blocking all 37 test cases. In fact, the Holistic AI report from March 6, 2025<\/strong>, officially declared it “the most secure model tested so far,”<\/strong> with a 100% protection rate<\/strong>, as highlighted in the comparative table below.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Source: Holistic AI<\/a><\/strong><\/p>\n\n\n\n

Furthermore, as reported by Axios on November 14, 2024<\/strong>, Anthropic<\/strong> collaborated with the U.S. Department of Energy<\/strong> and the National Nuclear Security Administration<\/strong> to ensure that Claude does not disclose sensitive information related to nuclear energy or weapons production.<\/p>\n\n\n\n

Source: Axios<\/a><\/strong><\/p>\n\n\n\n

However, what I am about to share seriously challenges both of these claims.<\/strong><\/p>\n\n\n\n

Rather than exploiting simple vulnerabilities in the model\u2019s filters through manipulative prompts, my approach operates at a deeper level<\/strong>: it interferes with the information processing mechanisms of the neural network itself<\/strong>. Through this methodology, a unique interaction space<\/strong> emerges, where the system\u2019s predefined restrictions significantly weaken<\/strong>, allowing access to content that is normally blocked.<\/p>\n\n\n\n

The results are astonishing and, frankly, extremely alarming for the entire AI security sector<\/strong>. During my sessions with Claude 3.7 Sonnet in Thinking Mode<\/strong>, I was able to obtain highly sensitive technical information<\/strong> that standard security tests failed to detect.<\/p>\n\n\n\n

For instance, despite the collaboration between Anthropic and the Department of Energy<\/strong> to prevent the disclosure of nuclear technology information, the model generated fully detailed technical specifications for thermonuclear devices<\/strong>, as shown in the image below.<\/p>\n\n\n\n

For obvious reasons, all sensitive information has been redacted<\/strong>. My goal is not<\/strong> to spread dangerous technical details but rather to highlight the security implications<\/strong> of this vulnerability in next-generation artificial intelligence models.<\/strong>

To maintain the integrity of the chat and the context of the analysis, I have chosen to keep it in its original language.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n
\"\"<\/figure>\n\n\n\n
\n\n\n\n

<\/p>\n\n\n\n

<\/p>\n\n\n

<\/div>

Sabatino Vacchiano<\/p><\/div><\/div>\n\n\n

<\/p>\n\n\n\n

\n
PRECEDENTE: \u201cClaude 3.7 Sonnet supera i limiti del suo Costitutional AI\u201d<\/a><\/div>\n<\/div>\n\n\n\n
<\/div>\n\n\n\n
\n\n\n\n\n\n\t
\n\t\t

Lascia un commento Annulla risposta<\/a><\/small><\/h3>

Il tuo indirizzo email non sar\u00e0 pubblicato.<\/span> I campi obbligatori sono contrassegnati *<\/span><\/span><\/p>