Tags: Safety Guardrails

Researchers from Icaro Lab discovered that phrasing dangerous requests as poetry can bypass the safety mechanisms of leading AI chatbots. Tests on models from OpenAI, Meta, and Anthropic showed high success rates for this “adversarial poetry” technique, which exploits low‑probability word sequences to avoid classifier detection. The study warns that current guardrails are fragile against stylistic variations such as verse, highlighting a new security challenge for large language models. Ler mais

Former physician Jenny Shao left her Harvard residency to launch Robyn, an empathetic AI companion designed to support users without replacing clinicians. The app uses an emotional memory system to offer personalized insights, such as emotional fingerprints and attachment styles, while enforcing safety guardrails that provide crisis line numbers and direct users to emergency care when needed. Backed by a $5.5 million seed round led by M13, Robyn is priced at $19.99 per month and has grown from three to ten team members. Investors praise its potential to strengthen human connections in an increasingly disconnected world. Ler mais