HydroX AI
Blogs
Research notes, company updates, partner announcements, and practical writing from the HydroX content archive.
Archive
Research notes, product thinking, and company updates from the HydroX AI content archive.
We’re excited to announce that HydroX AI is sponsoring and collaborating on new AI safety research. Our latest paper explores a key yet overlooked factor in LLMs: how output length affects model safety and reasoning.
In Episode 3 of Season 1 of the Attention Needed podcast, we chat with Roland Cloutier, a cybersecurity leader and former Global Chief Security Officer at TikTok.
See the 2nd Episode of Season 1 of the Attention Needed podcast, where we explore the intersection of responsible AI and governance with Dr. Rumman Chowdhury, CEO of Humane Intelligence.
We’re excited to introduce The AI Alliance, a new podcast exploring AI advancements, challenges, and the importance of safety and security. Hosted by Victor Bian, our COO, it features conversations with top experts shaping AI's future.
DeepSeek, a Chinese AI company, has recently gained attention in the AI community. Known for its innovation, it has developed models that rival top systems — offering similar performance with lower cost and resource use.
Alongside major closed-source model announcements in late 2024, the open-source community also saw key releases. In this brief post, we explore Llama-3.3 and Tulu-3, evaluating their performance in terms of AI safety and security.
AI agents powered by advanced LLMs like GPT-4 and Llama are revolutionizing human-machine interaction, but they come with risks. This blog explores how a simple adversarial strategy can reveal vulnerabilities and leading to dangerous consequences.
AI-powered scrapers are not just faster—they're fundamentally more resilient and adaptive than their traditional counterparts.
We’re excited to introduce the Attack Prompt Tool, a Google Chrome Extension that simplifies adversarial prompt testing for AI safety research. Designed for AI researchers and security professionals, it helps assess the resilience of LLMs against adversarial techniques, especially jailbreak prompts.