The News
Researchers from Google DeepMind and several universities demonstrated that ChatGPT can inadvertently reveal personal data through a simple ciberattack, costing only $200. This method involves instructing the model to repeat a specific word, leading to the extraction of sensitive information.
Why It Matters
This incident highlights significant vulnerabilities in AI models regarding data privacy and security. The ability to extract personal data raises concerns about the ethical implications of AI training processes and the need for stricter safeguards in AGI development.
Key Evidence
The article is sourced from cope, which reports on the findings of credible researchers from reputable institutions, indicating a reliable basis for the claims made.
Original Article
cope
ES source 路 Published Nov 30, 2023
https://www.cope.es/actualidad/tecnologia/noticias/ataque-algo-tonto-hace-que-chatgpt-revele-datos-personales-reales-procedentes-entrenamiento-20231130_3028915