Co-founder Benjamin Mann says the dual-use nature of powerful AI models means the company will not fully step back from monitoring how customers use them
Anthropic wants Singapore’s business. It does not want to stop watching how its most powerful models get used.
Co-founder Benjamin Mann said Wednesday that the Claude developer is unwilling to give up all oversight of customer usage. He cited the dual-use nature of advanced AI, meaning the same model that drafts a policy memo could, in the wrong hands, help with something far less benign.
The comments landed during Anthropic’s Singapore launch. That timing matters, because governments across the region increasingly want AI providers to keep sensitive data close to home and out of a vendor’s line of sight.
The event took place on October 14, 2026, during Singapore International Cyber Week. Anthropic opened its fifth Asia-Pacific office there, joining existing outposts in Tokyo, Bengaluru, Seoul, and Sydney.
The company also signed a Memorandum of Understanding with Singapore’s Ministry of Digital Development and Information, or MDDI. Mann signed for Anthropic, and MDDI Permanent Secretary Kai Fong Chng signed for the ministry.
Minister Josephine Teo witnessed the signing. So did Michael Sellitto, who heads policy for Anthropic in the Asia-Pacific region.
The agreement covers AI safety, security, model evaluation, and incident response. It also addresses responsible AI adoption in both the public and private sectors.
In practical terms, the MOU includes technical exchanges on AI testing standards. It also commits the two sides to supporting AI adoption inside government.
Then came the question that tends to follow any AI company into a sovereign-minded market. Fortune asked Mann whether Anthropic would offer local data residency and sovereign AI options as governments and companies push for more control over sensitive information.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
Mann’s answer was warm on ambition and careful on specifics.
Anthropic wanted customers “to run their most sensitive workloads with us.”
Singapore is not a speculative bet for Anthropic. The city-state ranks 2nd out of 121 countries in per-capita usage of Claude.ai, which is nearly six times the expected rate.
Government adoption is already deep. More than 1,500 engineers at GovTech Singapore use Claude Code, Anthropic’s coding tool.
Private-sector uptake is visible too. Grab and DBS are among the companies using Claude, and Grab’s BriX assistant reportedly saves about 40,000 hours each quarter.
On the data question, Anthropic has a partial answer already in place. It currently offers in-region inference for Claude Sonnet 5 through Amazon Bedrock, so prompts and responses stay inside the country.
Anthropic also plans to establish local data residency for its models in Singapore, though that remains a stated plan rather than a finished product.
The relationship has history. Anthropic’s collaborations with Singapore date back to 2023.
There is also a financial tie. GIC, Singapore’s sovereign wealth fund, invested in Anthropic during its 2025 and 2026 funding rounds.
Anthropic frames its approach as combining capability with safety. The company stresses that a model’s full capabilities may not be completely understood even after training is finished.
Anthropic’s model sits somewhere in between. Data can stay local through in-region inference, but the company retains some ability to oversee usage of its most powerful systems.
Singapore appears comfortable with that compromise, at least for now. The MOU leans into shared work on testing, evaluation, and incident response rather than a hands-off handover.
For enterprise and government buyers, the message is clear. Anthropic will compete hard for sensitive workloads, but customers should not expect a fully unsupervised, keys-to-the-kingdom deployment of its frontier models.
The remaining items to watch are when the planned local data residency actually arrives, and how much oversight Anthropic keeps once it does.