OpenAI signs pact to sell AI models to US defence agencies through Amazon cloud arm

OpenAI has signed a new agreement to provide its AI models to the US defence and government sectors via Amazon’s cloud division for both classified and unclassified projects, the company announced on Tuesday.
This contract allows OpenAI to support the Pentagon following a deal reached late last month, after the agency moved away from its former provider, Anthropic.
Anthropic, which secured a Pentagon contract valued at up to $200 million in July 2025, previously served as a primary defence AI supplier. In collaboration with Palantir and Amazon Web Services (AWS), it deployed Claude models within classified military and intelligence frameworks. However, that partnership dissolved in February when Anthropic restricted military applications of its AI, specifically regarding autonomous weaponry and domestic surveillance. Consequently, the Pentagon designated the firm a "supply chain risk," effectively halting its government projects.
While OpenAI previously focused on unclassified federal work, it has now secured the mandate for classified operations. Its AWS partnership highlights how defense contracts — facilitated by established cloud providers — have become a critical competitive arena. Success in the high-stakes public sector may also bolster OpenAI's appeal to corporate clients as a benchmark for reliability.
Bloomberg reported that the Pentagon is currently developing alternatives to Anthropic’s tools. Cameron Stanley, the Pentagon’s chief digital and AI officer, noted that transitioning from Anthropic products used in US operations in Iran would take over a month, though efforts to integrate a new large-language model are underway.
“The Department is actively pursuing multiple LLMs into the appropriate government-owned environments,” Stanley said in an interview.
“Engineering work has begun on these LLMs and we expect to have them available for operational use very soon,” Stanley added.
Following its shift to a for-profit model last autumn, OpenAI revised its Microsoft agreement to permit partnerships with rival cloud firms for national security clients.
Recently, both OpenAI and Elon Musk’s xAI received approval for classified Pentagon work. It remains uncertain how rapidly these tools can be integrated into existing systems like Palantir’s Maven Smart System, currently utilised in the Iran campaign, reported Bloomberg.
Until this shift, Anthropic was the sole AI system functional within the Pentagon’s classified cloud, where its Claude tool had gained significant popularity for its user-friendly interface.
Emil Michael, the under secretary of defence for research and engineering, said Alphabet Inc.’s Google is introducing its Gemini AI agents across the Pentagon’s three million-strong workforce to automate routine tasks. The company will initially operate on unclassified networks and then move into classified work, according to Bloomberg.
Michael said in an interview hours after Anthropic filed its lawsuit that the military was moving on and that there was little chance to revive talks.
For about a decade, Livemint—News Desk has been a credible source for authentic and timely news, and well-researched analysis on national news, busine...Read More
Catch all the Business News, Market News, Breaking News Events and Latest News Updates on Live Mint. Download The Mint News App to get Daily Market Updates.

Facts Only

* OpenAI has signed an agreement to provide its AI models to the US defense and government sectors.
* The agreement is facilitated by Amazon’s cloud division.
* The Pentagon moved away from Anthropic following a deal reached late last month.
* Anthropic previously secured a Pentagon contract valued at up to $200 million in 2025.
* Anthropic restricted military applications of its AI, specifically regarding autonomous weaponry and domestic surveillance.
* The Pentagon designated Anthropic a "supply chain risk."
* OpenAI has now secured the mandate for classified operations.
* The Pentagon is currently developing alternatives to Anthropic’s tools.
* Transitioning from Anthropic products used in US operations in Iran would take over a month.
* Google is introducing its Gemini AI agents across the Pentagon’s workforce.
* Emil Michael, the Under Secretary of Defence for Research and Engineering, noted Alphabet Inc.’s Google is operating on unclassified networks initially.

Executive Summary

The Pentagon is rapidly shifting its approach to artificial intelligence for defense applications following the termination of its partnership with Anthropic. The move underscores the significant risks associated with relying on a single AI provider, particularly in classified environments. The Department of Defense's need for reliable and secure AI solutions is accelerating, prompting a diversification of partnerships, including with OpenAI and Google. The timeline for transitioning away from Anthropic's Claude system is substantial, estimated at over a month for operations in Iran, highlighting the complexity and potential delays involved in integrating new technologies. While OpenAI’s entry into classified work represents a significant development, it also highlights the ongoing competition among AI providers seeking government contracts and the importance of cloud infrastructure in facilitating these partnerships. The Pentagon’s current pursuit of multiple Large Language Models (LLMs) demonstrates a pragmatic approach to mitigating risk and maximizing technological options.

Full Take

The shift from Anthropic to OpenAI represents a classic “motte-and-bailey” maneuver—the Pentagon, seeking a functioning AI solution after a perceived failure, has now embraced OpenAI, framing it as a pragmatic, responsive alternative. The underlying assumption is that OpenAI’s for-profit model, coupled with its expanded partnership network, provides a greater degree of resilience and flexibility than Anthropic’s previous, more narrowly focused approach. The admitted month-long delay in transitioning away from Anthropic’s Claude highlights a fundamental vulnerability: reliance on a single technology platform, particularly in a strategically sensitive domain like military intelligence. This situation also triggers a deeper pattern – the cyclical nature of technological over-reliance followed by panicked scrambling for replacements. The fact that both OpenAI and xAI received approval for classified work reinforces this pattern, indicating a wider competitive landscape spurred by the Pentagon’s needs. A significant, unspoken implication is the increasing concentration of power within cloud providers like AWS and Google, who now effectively act as gatekeepers to national security applications. The reference to Cameron Stanley's comment ("The Department is actively pursuing multiple LLMs") suggests a recognition of this power dynamic, but simultaneously risks replicating the vulnerabilities that initially prompted the change. Finally, the invocation of Google's Gemini operating across the entire Pentagon workforce speaks to a broader trend—the embedding of AI not just as a tool, but as an operational core – a potentially transformative, yet profoundly unsettling, development.
Patterns detected: ARC-0043 Motte-and-Bailey, ARC-0024 Ambiguity, ARC-0018 Systemic – cyclical dependence, ARC-0031 Authority Games – appeal to corporate partners.