A leading US intelligence official has suggested that to keep up with foreign adversaries, intelligence agencies should use commercially available AI while addressing privacy risks and broader concerns about the potential misuse of rapidly advancing technology.

Gilbert Herrera, the US Director of Research at the National Security Agency (NSA), stated in an interview that the intelligence community must find a way to capitalize on large AI models without infringing on privacy. “To fully harness the power of artificial intelligence for various applications, we’ll likely need to rely on partnerships with the industry,” he added.

Herrera expressed his desire for the NSA to utilize large commercial AI models trained on the open internet, referencing companies with access to vast amounts of data, such as Meta’s Facebook, Alphabet Inc.’s Google, and Microsoft Corp., which also owns GitHub, a platform used by software developers to store code.

Herrera acknowledged the risk of importing potentially biased algorithms into classified spying missions when using commercially available AI models. However, he emphasized that intelligence requirements can be met without accessing the underlying data of US citizens and companies used to train and develop the models.

Speaking at a summit on the modern conflict at Vanderbilt University, he explained that if a model trained on global data were used, access to the data wouldn’t be necessary; only the decision trees providing the desired information would be needed.

The National Security Council and the Office of the Director of National Intelligence did not immediately respond to requests for comment.

Herrera’s warning followed a meeting between Vice President Kamala Harris and CEOs of Alphabet, Microsoft, OpenAI Inc., and Anthropic. Harris stated that the White House supports efforts to minimize potential harm from AI technology. Harris emphasized the need for businesses to collaborate with the government to ensure safeguards that protect civil rights and privacy and prevent disinformation and fraud.

Wall Street firms are increasingly adopting AI technology. JPMorgan Chase & Co. recently introduced an AI-powered model to interpret Federal Reserve messaging and identify potential trading signals.

Although relying on commercial AI might be viable, Herrera’s suggestion could challenge the 2020 guidelines from the Office of the Director of National Intelligence that require US spy agencies to “maintain accountability for iterations, versions, and changes” made to any AI model and to identify and mitigate “undesired bias.”

Herrera cautioned that the US must find a solution to this problem, as adversaries will exploit commercial AI. He noted that those using large data sets would have an advantage, citing China’s extensive video surveillance and other data sources. China aims to become the world leader in AI by 2030.

Herrera asserted that the intelligence community must address the issue of using publicly trained information or risk being unable to utilize specific AI capabilities, allowing adversaries to exploit the technology entirely. He also mentioned that private companies’ algorithmic models could benefit the NSA’s work on coding and report writing for classified information assessment.

Herrera emphasized the importance of respecting civil liberties and privacy, acknowledging the complexity of the problem.

Integration of commercially available AI models into the operations of intelligence agencies presents several unique advantages and challenges that must be considered, viz:

• Utilizing private AI models enables intelligence agencies to take advantage of the industry’s accelerated rate of innovation. As AI development accelerates, the private sector is becoming more agile and receptive to emerging trends and technologies, which could help intelligence agencies remain ahead of their opponents.
• Intelligence agencies can reduce costs associated with developing and maintaining their AI systems using commercially available AI models. This enables them to allocate resources and concentrate on their core missions more effectively.
• Collaboration on talent: Through partnerships with private AI companies, intelligence agencies can access top talent in the field. This partnership could advance AI-driven intelligence collection, analysis, and decision-making.
• While the use of commercial AI models can provide numerous benefits, it also raises ethical issues. In addition to possible biases in the algorithms, there may be concerns regarding using privately-developed AI for national security purposes and the potential for misapplication or abuse of these potent technologies.
• Transparency and accountability: Intelligence agencies must ensure that their AI-powered processes are transparent and accountable when utilizing commercial AI models. Establishing clear guidelines and oversight mechanisms can help ensure that artificial intelligence is used responsibly and ethically while addressing potential biases and other undesirable outcomes.
• As intelligence agencies rely more on AI models that are commercially available, they must prioritize data security to safeguard sensitive information from potential breaches. To ensure data security and privacy, this may necessitate the development of stringent data protection protocols and close collaboration with private AI companies.
• Given the global character of AI development and today’s interconnectedness, intelligence agencies could benefit from increased international cooperation in AI research and development. This may involve exchanging best practices, addressing ethical concerns jointly, and cooperating to mitigate potential risks AI technology poses.

In conclusion, while adopting commercially available AI models by intelligence agencies can provide substantial benefits, addressing the unique challenges and concerns is necessary. Intelligence agencies can harness the power of AI to improve their operations and better safeguard national interests by navigating these issues with care.