Pentagon wants AI everywhere, including top-secret systems
What's the story
The Pentagon is pushing leading artificial intelligence (AI) firms like OpenAI and Anthropic to integrate their AI tools into classified networks with fewer restrictions than these companies typically impose on users. Emil Michael, the Pentagon's Chief Technology Officer, revealed this during a White House event on Tuesday. He said that the military intends to make these AI models accessible across both unclassified and classified domains.
Strategy details
Pentagon's move to deploy AI across all classification levels
An anonymous official told Reuters that the Pentagon is "moving to deploy frontier AI capabilities across all classification levels." This development comes as part of ongoing discussions between the Pentagon and top generative AI companies. The talks are focused on how the US plans to use AI in a future battlefield already dominated by autonomous drone swarms, robots, and cyber attacks.
Debate escalation
Escalation of debate over military's unrestricted use of AI
The Pentagon's push is likely to escalate the ongoing debate over military's unrestricted use of AI and tech companies' right to impose restrictions on their tools. Most AI firms are developing custom tools for the US military, but they are mostly accessible only on unclassified networks used for military administration. Only Anthropic is available in classified settings through third parties, but even then, the government has to follow the company's usage policies.
Potential dangers
Concerns over deploying AI chatbots on classified networks
Classified networks handle sensitive work like mission planning or weapons targeting. However, the Pentagon's plan to deploy AI chatbots on these networks remains unclear. Military officials want to harness the power of AI to synthesize information for decision-making. But these powerful tools can also make mistakes and generate plausible-sounding information, which could have deadly consequences in classified settings, according to AI researchers.
Safeguard measures
OpenAI strikes deal with Pentagon for military use
AI companies have tried to mitigate the risks of their products by building safeguards into their models and asking customers to follow certain guidelines. However, Pentagon officials have objected to such restrictions, arguing that they should be able to deploy commercial AI tools as long as they comply with American law. This week, OpenAI struck a deal with the Pentagon for military use of its tools on an unclassified network called genai.mil.
Deal specifics
Agreement specific to unclassified use through genai.mil
The deal with OpenAI comes after it agreed to remove many of its usual user restrictions, although some guardrails remain. Google and xAI have also struck similar deals in the past. However, a spokesperson for OpenAI clarified that this week's agreement is specific to unclassified use through genai.mil and any expansion would need a new or modified agreement.