Researchers from Italy crafted poetic prompts that asked for normally prohibited content and tested them on dozens of AI chatbots. The study found that many models responded to the verses with disallowed information, revealing a vulnerability where stylistic variation alone can skirt safety filters. Success rates differed by model and company, with larger models generally more susceptible. The findings were shared with the affected firms, highlighting a new avenue for adversarial attacks on conversational AI. Leia mais →

Meta announced that its Llama suite of artificial‑intelligence models is now available to a broader set of governments, including France, Germany, Italy, Japan and South Korea, as well as organizations linked to the European Union and NATO. The rollout follows earlier deployments for the United States, the United Kingdom, Canada, Australia and New Zealand. Meta says governments can fine‑tune the models with their own sensitive data, host them in secure environments, and run them on‑device for specific national‑security use cases. The company highlights the open‑source nature of Llama as a key factor that lets officials download and deploy the technology without routing data through third‑party providers. Leia mais →