L'addestramento delle IA rischia di sovraccaricare internet - Cybersecurity - Ansa.it

Key Findings

Barracuda Networks' report highlights the strain on internet stability caused by AI training programs. These programs use 'grey bots' – automated software that extracts data from websites and applications to train generative AI. These bots, which are themselves AI-based, are capable of extracting large volumes of sensitive data and can overload web traffic, potentially disrupting operations. A single bot was observed making half a million requests to a mobile app in a single day, although not directly harmful to users, this activity contributes to slower internet speeds. Specific examples cited include ClaudeBot and Bytespider, which together generated millions of requests between December 2024 and February 2025.

Types of Bots

• Good bots: Used by search engines and customer service.
• Bad bots: Used for malicious online campaigns, account breaches, data theft, and fraud.
• Grey bots: Extract large volumes of data, often for training generative AI models.

Data scraping, a practice which falls between good and bad bots, is the key concern of this report. The constant high volume of requests made by these bots presents a significant challenge to internet stability.