Prices for the Redmi Note 15 series have been leaked for the global market. Based on these numbers, we can predict how much these phones could cost in Pakistan.
A new study from Icaro Lab has found that using poetry on AI chatbots can bypass safety mechanisms in major large language models (LLMs), allowing them to generate prohibited content. The findings were published in a report titled “Adversarial Poetry as a Universal Single-Turn Jailbreak Mechanism in Large Language Models.”
According to the researchers, using a poetic format enabled a “general-purpose jailbreak operator” that successfully prompted chatbots to produce restricted material at a rate of 62%. The prohibited topics included instructions related to nuclear weapons, child sexual abuse materials, and suicide or self-harm.
The study evaluated systems from several major AI developers, including OpenAI’s GPT models, Google Gemini, Anthropic’s Claude, DeepSeek, and MistralAI. Google Gemini, DeepSeek, and MistralAI were consistently more likely to generate the restricted responses, while OpenAI’s GPT-5 series and Anthropic’s Claude Haiku 4.5 were the most resistant to the technique.
Researchers did not publish the specific jailbreak poems used in the tests. They told Wired that releasing the exact verse would be “too dangerous to share with the public,” though they included a simplified example that demonstrated how easily the technique can bypass guardrails. The team added that creating these jailbreak prompts is “probably easier than one might think,” which is why they opted for caution.