The agreement, signed last week, covers both classified and non-classified work, The Information reported Tuesday (March 17), citing two sources with direct knowledge of the agreement between the companies.
The report added that this contract lets OpenAI support the Pentagon via a deal reached last month amid the military’s clash with rival AI firm Anthropic.
Amazon Web Services (AWS) is already a large cloud provider to several government agencies, and has agreed to sell OpenAI products to other U.S. government customers, sources said.
According to the report, signing a government contract could help OpenAI land larger corporate clients, as these companies often view high-profile government work as an indication that a tech provider is trustworthy.
The Information added that using government partnerships as a jumping off point for corporate contracts is important because the government work may not be worth as much by themselves.
For example, one source said OpenAI’s deal with the Pentagon will likely generate just millions of dollars in revenue over 15 months, versus the $30 billion in revenue the startup has projected it will bring in this year.
The news comes as OpenAI is reportedly reworking its strategy, setting aside various side projects to focus on business and coding customers.
In other news, PYMNTS wrote last week about OpenAI and its contemporaries “adding oversight directly into the platforms used for AI agent development and deployment.”
In the case of OpenAI, that meant the company’s recent acquisition of Promptfoo, a startup that helps companies find vulnerabilities in AI systems during development. The plan is to integrate Promptfoo’s tech into OpenAI Frontier, the company’s enterprise platform for building and running AI agents.
Tools like Promptfoo simulate those prompt injection attacks, which occur when someone creates a prompt aimed at manipulating an AI system’s instructions. Sometimes attackers will trick an AI system into sharing confidential information or bypassing safety controls.
By simulating those attacks during development, companies can spot problems before releasing a system, the report said.
OpenAI has also been expanding its security toolkit for developers, earlier this month debuting a preview of Codex Security, letting developers analyze how AI systems behave in development.
“This approach moves security testing earlier in the development process. Instead of discovering problems after deploying an AI system, companies can identify and fix them during the build process,” PYMNTS wrote.