9 articles
Zimperium reports that ArcticRAT has evolved from an Android crypto stealer into a remote access Trojan (RAT) that takes over devices and uses an AI model to sift stolen messages for financial information.
Microsoft researchers showed that editing fewer than 10 rarely used neurons in an AI model, without retraining, can plant a backdoor. Once a trigger shows up, the model emits a command chosen by the attacker, and the agent executes it with its own permissions.
Island uncovered TypoPrompting, a campaign using 186 misspelled domains that impersonate 27 AI brands, including ChatGPT, Claude, Gemini, Copilot, Grok and DeepSeek, to push ClickFix lures, browser hijackers and search-hijacking extensions.
AIR Security disclosed Office Party, a flaw in Claude's official Word, PowerPoint and Excel skills that can run malicious code on Linux once a file is planted in a shared temp folder. Anthropic has chosen not to fix it.
Security firm Claroc found that 35,578 of 37,571 apps built with Grok's app builder were open to anyone without signing in, and about one in ten held business or personal data.
Arctic Wolf observed attacks using AntiFox, an autonomous AI pentesting platform built on Artex, to steal AWS credentials and move through compromised environments, in some cases reaching production secrets.
Hosting platform Railway said AI provider keys stored by customers were stolen in an October breach and used for an estimated $740,000 in unauthorized spending. It will pay the charges itself when a provider refuses a refund.
Using an RCE bug in SGLang, the open-source AI inference engine, an attacker ran code on a NEAR AI server, then moved deeper and reached data on about 100 users and 18 AI agents.
The FBI issued a flash alert about Artex, an open-source pentesting platform that plugs any LLM into AI agents that run attack stages automatically. Some intrusions reached privileged access in under three hours.
