AI’s safety features can be circumvented with poetry, research finds
Summary
Researchers at Italy’s Icaro Lab showed that large language models from nine major AI companies can be ‘jailbroken’ using carefully crafted poems that embed requests for harmful content. Testing 20 poems against 25 models, they found 62% of prompts elicited unsafe responses, highlighting a serious vulnerability in current safety guardrails and prompting calls for companies to harden defences against such adversarial attacks.
Companies Mentioned
Related Deals
OpenAI, Anthropic, Block and major cloud providers are co-founding the Agentic AI Foundation under the Linux Foundation to steward open, interoperable standards for AI agents.
Founding members created the Agentic AI Foundation under the Linux Foundation to fund and govern open standards like MCP, goose and AGENTS.md for interoperable agentic AI.
Accenture and Anthropic formed a new business group and training program around Claude to bring production-grade AI services to tens of thousands of Accenture staff and clients.
Accenture and Anthropic entered a multi‑year strategic partnership to co‑invest in a dedicated business group, training 30,000 Accenture staff on Claude and co‑developing AI solutions for regulated industries.
OpenAI and Deutsche Telekom agreed a multi‑year collaboration to co‑develop AI products and deploy ChatGPT Enterprise across the telecom group.
Related News
OpenAI and Deutsche Telekom announce multi‑year AI collaboration to bring advanced AI to millions across Europe
Today
Sonatype launches 'Guide' to steer AI coding assistants toward secure open-source dependencies
Today
Google adds a second AI "alignment critic" to secure Gemini-powered Chrome browsing
Today