Pentagon’s new plan: Artificial intelligence models will be trained with secret information

It has been claimed that the US Department of Defense is making remarkable preparations to take artificial intelligence technologies to a further level for military purposes. According to information provided by MIT Technology Review, the Pentagon plans to collaborate with artificial intelligence companies to train models directly on secret military data. This approach coincides with the US’s goal of establishing an “AI-first” structure in the field of defense. This transformation was clearly expressed in the strategic document published by Defense Secretary Pete Hegseth earlier in the year.

It is known that the US army currently uses various artificial intelligence tools. For example, allegations that the Claude model developed by Anthropic was used in operations to capture Venezuelan President Nicolás Maduro and in military activities against Iran were reflected in the public. Despite this, it was noteworthy that the technology in question was later banned for federal institutions. However, it is stated that models that will be trained directly with hidden data have the potential to produce much more specific and context-sensitive results. In particular, data on past operations that are not publicly available can significantly increase the analysis capacity of these systems.

The Pentagon will retain ownership of all data produced

According to the plans, this training process will be carried out only in high-security data centers authorized to host secret state projects. It is stated that the Pentagon will retain ownership of all data used in this process by training special versions of artificial intelligence models. Despite this, it is also said that in rare cases, some employees of artificial intelligence companies may gain limited access by obtaining the necessary security permissions.

On the other hand, this approach brings with it significant risks. Aalok Mehta, who previously worked on artificial intelligence policies at Google and OpenAI, states that model training with confidential data may lead to different security vulnerabilities, even if it does not directly mean data leakage. Especially if the same model is used widely throughout the Ministry of Defence, it is noteworthy that personnel with low authority levels may indirectly access information that they should not have access to. This situation brings up new discussions in terms of military hierarchy and information security.

In addition, it has begun to become clear which technology companies the Pentagon will work with within the scope of this initiative. While it is known that OpenAI and Elon Musk’s initiative xAI signed an agreement with the Ministry of Defense, it is stated that Anthropic, which has been cooperating with the state for a long time, may not take part in the project. The company’s refusal to be used in mass surveillance and autonomous weapon systems stands out as one of the main reasons behind this decision. Despite this, as the role of artificial intelligence in the field of defense gradually expands, the scope of cooperation between technology companies and public institutions is also being reshaped.

In addition to all this, global discussions on the use of artificial intelligence in the military field have also gained momentum. Organizations such as the European Union and the United Nations argue that stricter regulations should be introduced on the use of autonomous weapon systems and artificial intelligence in the war environment.