Transaction

jmetaB02f17e01dca51c5883aaaabca9f624a67b7ad98164f1fc6add3fc3e234c60f9aef@104e04f4dc7bbb58b675a0be8ec8a2392cd828cadc0c1b85347e2d4ab003150erss.itemmetarss.netMŸ<item> <title>Synthetic Cancer -- Augmenting Worms with LLMs</title> <link>https://arxiv.org/abs/2406.19570</link> <description>arXiv:2406.19570v1 Announce Type: new Abstract: With increasingly sophisticated large language models (LLMs), the potential for abuse rises drastically. As a submission to the Swiss AI Safety Prize, we present a novel type of metamorphic malware leveraging LLMs for two key processes. First, LLMs are used for automatic code rewriting to evade signature-based detection by antimalware programs. The malware then spreads its copies via email by utilizing an LLM to socially engineer email replies to encourage recipients to execute the attached malware. Our submission includes a functional minimal prototype, highlighting the risks that LLMs pose for cybersecurity and underscoring the need for further research into intelligent malware.</description> <guid isPermaLink="false">oai:arXiv.org:2406.19570v1</guid> <category>cs.CR</category> <category>cs.AI</category> <arxiv:announce_type>new</arxiv:announce_type> <dc:rights>http://creativecommons.org/licenses/by-nc-nd/4.0/</dc:rights> <dc:creator>Benjamin Zimmerman, David Zollikofer</dc:creator> </item>