Microsoft releases generative AI for spies that can reduce hacking risk without internet connection



Many generative AIs, including OpenAI's chat AI 'ChatGPT', use cloud services to learn and infer patterns from data. Therefore, existing generative AIs are always at risk of data leakage and hacking. On May 2, 2024, Microsoft released a generative AI that is completely disconnected from the Internet for American intelligence agencies.

Microsoft Creates Top Secret Generative AI Service for US Spies - Bloomberg

https://www.bloomberg.com/news/articles/2024-05-07/microsoft-creates-top-secret-generative-ai-service-for-us-spies



Microsoft launches AI chatbot for spies | Ars Technica

https://arstechnica.com/information-technology/2024/05/microsoft-launches-ai-chatbot-for-spies/

Microsoft has been working on developing the system for the past 18 months by adapting an existing AI supercomputer in Iowa.

The generative AI models that Microsoft has deployed for intelligence agencies can read files provided by users but cannot connect to the open internet, and can also perform top-secret analysis in chat-style conversations like ChatGPT and Microsoft Copilot.

By not connecting to the internet, you can reduce the risk of confidential data being compromised or hacked.



William Chappell, Chief Technology Officer, Strategic Mission and Technology at Microsoft, said about this generative AI, 'We deployed the key elements that support the GPT-4-based generative AI in a cloud with an air-gapped environment isolated from the Internet.' According to Chappell, this generative AI is built on a special network that is only accessible to the US government, and although the stored information can be read, it cannot be used to train the AI.

'Governments can keep generative AI models clean and prevent confidential information from being learned by other AI platforms,' said Chappell. 'You don't want the AI to learn the questions you ask and then somehow reveal the information it learns.' In theory, about 10,000 people could have access to the generative AI.



At the time of writing, the generative AI is undergoing testing and certification by several intelligence agencies, including the CIA. 'It's already in production and deployed at some intelligence agencies,' Chappell reports.

in Software, Posted by log1r_ut