Friday, July 12, 2024
- Advertisement -

    Latest Posts

    Microsoft Deploys GenAI Model For US Intelligence Agencies: Here’s What’s Worrying

    In a first, Microsoft has provided United States intelligence agencies with a generative AI model, that’s isolated/disconnected from the internet, to enable analysis of classified and top-secret information, according to a report by Bloomberg.

    According to the report, Microsoft’s special product for the US spy agencies is a GPT 4-based model and is supported onto a cloud with an “air-gapped” environment that is isolated from the internet. This is to prevent risks that the sensitive information could leak into the open or potentially get hacked if intelligence agencies use large language models that rely on cloud services to examine data.

    The generative AI model is on a special network that’s only accessible by the US government, Microsoft’s Chief Technology Officer William Chappell said to Bloomberg. He further explained that the GPT-4 model placed in the cloud is static, i.e. it can read files, but not learn from them or the open internet. This allows the government to keep the model clean, preventing secret information from getting absorbed into the platform.

    What is GPT-4?

    GPT-4 is a generative AI language model developed by OpenAI and is trained on Microsoft Azure AI supercomputers. In 2023, Microsoft had announced greater investments in OpenAI’s research and development of AI systems in exchange for using such technology across its products. GPT-4 is considered to be “more reliable, creative, and able to handle much more nuanced instructions than GPT-3.5.”

    According to OpenAI, like previous models, GPT-4 has the tendency to “hallucinate” facts and make reasoning errors. MediaNama has reported earlier that it also lacks knowledge of events after September 2021 and does not learn from its experience. Further, it can sometimes make simple reasoning errors or be extremely gullible in accepting obvious false statements from a user. It can also fail at hard problems such as introducing security vulnerabilities into the code it produces. Another issue is that GPT-4 can be confidently wrong in its predictions as it does not double-check its work.

    Can the use of AI augment the surveillance capabilities of US spy agencies?

    The acquisition of a new AI tool for boosting spying operations of US agencies comes at a time when the US Senate is set to vote on the Reforming Intelligence and Securing America Act (RISAA) that allows surveillance of non-Americans without a warrant. The bill reauthorises section 702 of the Foreign Intelligence Surveillance Act (FISA) that allows the American government to conduct surveillance on foreign citizens through, “electronic communications service providers.” This means that US businesses providing services such as e-mails, phone calls, text messages, etc. are required to assist the government in intercepting communication.

    Additionally, the US military has also been using AI for the creation of new cybersecurity tools to secure critical infrastructure, for combat testing, research and for developing AI-powered autonomous vehicles among other purposes.  

    These developments raise concerns about the data of Indian citizens. The RISAA, if implemented, expands the scope of US surveillance over foreign citizens’ data. In addition to data accessed through electronic communication service providers, by now US AI companies like OpenAI have already accessed a large amount of publicly available data of users across the world. This adds to the data trove of US spy agencies to snoop on Indian citizens’ use of AI as well.

    As MediaNama Founder-Editor Nikhil Pahwa had pointed out earlier, generative AI can be used for analysing large datasets to detect and identify vulnerabilities and strategies for cyberattacks. It is also important to note that India’s data protection law does not govern publicly available personal data. This means publicly available personal data of Indian citizens can possibly be used for AI training, surveillance, and other purposes which can potentially impact people’s rights to privacy and security.

    Further, given the limitations of GPT4 discussed above, it’s not known whether there’s an oversight system to examine the output generated by the model. Notably, the US government’s latest memorandum ensuring accountability of use of AI in governance exempts law enforcement agencies and sensitive law enforcement information. Emphasising the growing reliance of US intelligence on AI for surveillance measures and for determining threat actors to national security, the American Civil Liberties Union stated in its response to the memo, that these agencies are not bound by specific rules or safeguards for their AI systems, as well as clear processes for enforcement of rules. It added that exceptions for “sensitive law enforcement” information may end up undermining the memo.

    Also Read:

    The post Microsoft Deploys GenAI Model For US Intelligence Agencies: Here’s What’s Worrying appeared first on MediaNama.

    Latest Posts

    - Advertisement -

    Don't Miss

    Stay in touch

    To be updated with all the latest news, offers and special announcements.