Subliminal Learning in AIs

Today’s freaky LLM behavior: We study subliminal learning, a surprising phenomenon where language models learn traits from model-generated data that is semantically unrelated to those traits. For example, a “student” model learns to prefer owls when trained on sequences of…

Bloomberg’s Comdb2 Vulnerable to DoS Through Malicious Packets

Cisco Talos’ Vulnerability Discovery & Research team has disclosed five critical security vulnerabilities in Bloomberg’s Comdb2 open-source database that could allow attackers to cause denial-of-service conditions through specially crafted network packets. The vulnerabilities, all affecting version 8.1 of the high-availability…

Tridium Niagara Framework Flaws Expose Sensitive Network Data

Cybersecurity researchers at Nozomi Networks Labs have discovered 13 critical vulnerabilities in Tridium’s widely-used Niagara Framework, potentially exposing sensitive network data across building management, industrial automation, and smart infrastructure systems worldwide. The vulnerabilities, consolidated into 10 distinct CVEs, could allow…