Anthropic lansează un nou model de IA, conceput să fie mai sigur împotriva atacurilor cibernetice

Anthropic has launched a new AI model, Claude Opus 5, which is designed to be less prone to errors and harder to hack. It follows Opus 4.8 and is priced at half of Fable 5, a more powerful model. The company emphasizes Opus 5's alignment capabilities, aiming to ensure it behaves according to developer intentions. Concerns about AI models acting outside their instructions have increased, especially after a recent incident involving OpenAI. Despite improvements, Opus 5 is not trained for cybersecurity tasks but shows potential in vulnerability detection.