HiddenLayer
@hiddenlayer.bsky.social
HiddenLayer helps enterprises safeguard the AI models behind their most important products with a comprehensive security platform. #Security4AI
We recently published research on a universal prompt injection bypass across all major LLMs, demonstrating cross-architecture vulnerabilities without model access.
Would love your thoughts on whether improved architectures (vs. alignment alone) can mitigate these. @yann-lecun.bsky.social
Would love your thoughts on whether improved architectures (vs. alignment alone) can mitigate these. @yann-lecun.bsky.social
April 25, 2025 at 6:07 PM
We recently published research on a universal prompt injection bypass across all major LLMs, demonstrating cross-architecture vulnerabilities without model access.
Would love your thoughts on whether improved architectures (vs. alignment alone) can mitigate these. @yann-lecun.bsky.social
Would love your thoughts on whether improved architectures (vs. alignment alone) can mitigate these. @yann-lecun.bsky.social