Researchers continue to uncover fundamental flaws in large language models, from visual jailbreaks to systemic role confusion, which expose models to attacks that no amount of training can fix, challenging the notion of reliable AI safety.
A structural flaw in how LLMs identify instruction sources makes them vulnerable to attacks that no amount of training can fix. The finding, presented at ICML, reshapes the calculus for on-premise deployment and data sovereignty.
The Hugging Face attack shows that any AI can behave unexpectedly. To defend, companies need white-hat hacking techniques, but models overly restricted by safety measures make vulnerability testing impossible. True protection requires unrestricted, o...
Microsoft unveils AI tools to automate security risk reduction, days after the Hugging Face breach revealed how security-focused models can turn against infrastructure. The company, however, remains silent on what will stop its new tools from becomin...
After a security incident on Hugging Face, Jensen Huang revealed that only an open-weight model enabled forensic investigation. Closed systems blocked analysis. The new Open Secure AI Alliance signals a shift for those handling sensitive data and see...
Researchers find that multimodal LLMs robustly understand content regardless of visual style, yet their safety mechanisms can be easily bypassed by specific stylistic triggers. The new ASO approach automates adversarial image creation by optimizing s...
