The US government is stepping up its efforts to integrate advanced models into the secret services, fearing that China will gain a strategic advantage, the Economist said in a statement. This concern was raised after the presentation of the Chinese Linguistic Deepseek model, a development that Donald Trump described as a “awakening bell”, while Senate Information Committee Vice President Mark Warner admitted that the US intelligence community was surprised.
In this context, the Biden government has instructed information services, the Pentagon and the Ministry of Energy to intensify experimentation with the most modern models. On July 14, the Pentagon assigned contracts of up to $ 200 million to companies such as Anthropic, Google, Openai and XAI, with the aim of developing ‘agentic’ models, which can execute autonomous complex commands. Already, specialized software versions, such as Anthropic’s Claude Gov, are used by security services, adapted for gradual data analysis and linguistic competence in strategic dialects.
. Katrina Maligan, a former executive of the Department of Defense and Information and current head of partnerships in Openai, notes that “the adoption of artificial intelligence in the field of national security is probably not yet at the desired level”. Torun Chambra, a former head of Political Technology at the National Security Council and today’s Anthropic official, points out that the actual transformation is not only about the use of a chatbot, but to “radically redesign the way the mission is executed”.
China
At the same time, concern for China remains strong. Analysts such as Philip Rainer of the Institute for Security and Technology estimate that Beijing, having “fewer moral barriers”, may utilize models faster and effectively.
Senator Warner criticizes, saying that US services are doing “miserable work” in monitoring Chinese progress. As Maligan summarizes, the greatest risk for the US is not the hasty adoption of technology. “For general artificial intelligence, but to lose the race for its adoption.”
Europe
Similar initiatives are being developed in Europe, the Economist points out. The United Kingdom provides access to classified linguistic models throughout its (UKIC) community, while the French Mistral, the only essential player in Epirus, works with the French artificial intelligence service, Amiad. Mistral’s SABA model is trained in data from the Middle East and South Asia, which makes it particularly effective in Arabic and local languages such as Tamil’s.
Despite progress, there are serious reservations. Richard Carter of the Alan Turing Institute stresses that secret services need “consistency, reliability and transparency”, features that today’s models do not fully guarantee, as they exhibit the phenomenon of “hallucinations” (incorrect answers).
He warns that agentic models, noting that Openai’s latest related model shows hallucinations in about 8% of the answers.
Source:
