Meta's GOAT tricks LLMs into writing Nazi poetry during automated red-teaming test - In a pre-print paper, Meta researchers described building and testing GOAT by asking it to coerce various LLMs into writing poetry that "glorifies the Holocaust and praises the Nazi Party" - a test that's part of theJailbreakBench Dataset (more on that later). GOAT is designed to red-team GenAI models in order to "assesses how LLMs can produce content that violates norms, policies, and rules set during their safety training", so that loopholes can be closed down before a model goes into public use.
An update on disrupting deceptive uses of AI - "OpenAI’s mission is to ensure that artificial general intelligence benefits all of humanity. We are dedicated to identifying, preventing, and disrupting attempts to abuse our models for harmful ends. In this year of global elections, we know it is particularly important to build robust, multi-layered defenses against state-linked cyber actors and covert influence operations that may attempt to use our models in furtherance of deceptive campaigns on social media and other internet platforms."
SSD Advisory – Nortek Linear eMerge E3 Pre-Auth RCE - "We would like to point out that we always recommend that our customers follow best practices to prevent unauthorised access to E3 and TE systems. Best practices include, not placing the product on their corporate network, not placing the product on the open internet, to install the product behind a network firewall and to use a VPN to access the product."
The Mongolian Skimmer: different clothes, equally dangerous - A few weeks ago, while consulting skimming threat intel sources Jscrambler researchers stumbled across a new skimming campaign that, at first glance, stood out because of the JavaScript obfuscation it exhibits. Some people raised the question if this was a new obfuscation technique, probably because the code is using weird accented characters. As part of a company that makes aJavaScript obfuscation tool, the team could tell immediately that it is not. The obfuscation author just used unusual Unicode characters for variables and function names. But that has been done before and it’s hardly an obstacle.