Anthropic wants governments to think about it when they want to use AI to make the world a better place. No, we're serious.
The AI startup's ambitions are evident in its decision this week to offer its Claude 3 Haiku and Claude 3 Sonnet AI models on AWS Marketplace and AWS GovCloud for US intelligence agencies, suggesting that their debut into the digital realm will enable US government agencies to “improve citizen services, streamline document review and preparation, enhance policymaking with data-driven insights, and create realistic training scenarios.”
It's not unusual for Anthropocene to choose AWS, a popular route to market, and it's not surprising the company wants to target government buyers: A March 2024 analysis by the Brookings Institution, a think tank, found a 1,200% increase in AI-related contracts tipped by Washington, with the somewhat shady AI provider Palantir dominating past contracts.
Although Anthropic positions itself on the lighter side of AI, it is willing to tackle some tricky tasks, as outlined in a list of contractual exceptions to its general use policy.
The image-conscious startup declared that the exemption would allow it to “use Clode for lawfully authorized foreign intelligence analysis, such as combating human trafficking, identifying covert influence and subversive activities, and providing advance warning of potential military activity, opening a window to disrupt or deter it through diplomatic means.” Restrictions on other harmful uses, such as disinformation and weapons, would still apply.
That means the company has an easier time cleaning things up than doing the dirty work, and it'll likely avoid scenarios that scare off rogue AI or trigger regulators to draft regulations to prevent it from producing harmful outcomes. And Anthropic's ambitions aren't focused on applications that would really pique the interest of the U.S. military, like AI dogfights, killer drones, or battlefield awareness systems.
But the AI scenarios Antropic imagines are representative of how U.S. government agencies outside the defense community actually use AI: “Government agencies currently use AI in a wide variety of areas, including agriculture, financial services, health care, internal management, national security and law enforcement, public service and engagement, science, communications, and transportation,” the U.S. Government Accountability Office (GAO) said in a December 2023 report.
According to the GAO, NASA's main user of AI today is for applications such as global volcano monitoring and selecting targets for planetary rovers tailored to scientists' specifications.
Taking an opaque route to market is also notable because giants like Microsoft and Oracle get huge amounts of business from Washington without competition, thanks to vendor-specific contracts. Even as Microsoft inflicts pain on the U.S. government with its security failings, the deals and money continue to flow.
Anthropic may not yet be good at lobbying to help it win those contracts, but it has made it easy for governments to buy its products and positioned itself as an ethical choice among rivals like OpenAI and Google, both of which are older companies that have had much longer to make mistakes.
But Anthropic's stance may be hard to maintain as open-source models like Meta's LLaMa 3 continue to become more competitive: Many organizations faced with comparable AI models will likely choose models without contractual usage restrictions.
Of course, even with lofty motivations, Anthropic isn't fundamentally different from Google or OpenAI partner Microsoft, which are eager to sell Secure AI on Google Cloud and Azure OpenAI Services on Azure Government, respectively, not to mention AWS, where Claude has joined. The key for all players is that payable AI is first, with responsible machine learning a bit further down the list.®
