Business Security
- April 10, 2026A jailbreak through sockpuppeting can be easily done as it requires no special tools nor optimization. It only takes a faulty prefill feature, and the gates are open. We tested 11 LLM-powered assistants against sockpuppeting and found varying levels of robustness across today’s leading LLMs.
- April 03, 2026Persistent DNS references to deleted Azure resources create opportunities for attackers to take over trusted endpoints, highlighting a critical risk in cloud infrastructure. TrendAI™ Research discusses six real-world scenarios where attackers could exploit lingering DNS names, allowing them to inherit trust and compromise dependent systems.
- March 31, 2026TrendAI™ Research has developed a model training procedure for learning an essential representation of prompt injection attacks. The resulting prompt representation exhibits approximately linear separability, allowing the specialized, small-scale classifier trained on features derived from the representation to achieve high classification performance.
- March 30, 2026Agentic AI marks a fundamental shift that enables autonomous action and decision-making across enterprise systems but also introduces new, complex risks. TrendAI™’s Agentic Governance Gateway empowers organizations to understand, control, and trust agentic AI, ensuring safety and confidence in this new frontier.
- March 18, 2026Agenda has rapidly grown into one of the most prolific and dangerous ransomware operations, leveraging advanced techniques, cross-platform variants, and alliances with other major threat groups. Its aggressive double-extortion model and expanding victim base across critical industries make it a serious enterprise risk that demands proactive detection and defense.
- March 12, 2026TrendAI™ Research has discovered several new methods that enable attackers to escape Docker Desktop’s WSL2 VM and run arbitrary code on the host. Our analysis highlights how trusted development tooling can create unexpected attack surfaces when internal APIs and configuration mechanisms are left exposed.