Anthropic Just Signed a Safety Pact With Singapore. Governments Are Becoming AI Infrastructure

On October 14, Anthropic opened its Singapore office and signed a memorandum of understanding with Singapore’s Ministry of Digital Development and Information (MDDI) covering AI safety, model evaluation, incident response, and responsible adoption. It is the company’s fifth office in Asia-Pacific, after Bengaluru, Seoul, Sydney, and Tokyo.

This is the second government-linked expansion in a week. On October 10, Anthropic announced its first India office in Bengaluru. On October 12, the UK regained access to safety-test US frontier models. The pattern is no longer a coincidence: frontier AI labs are embedding themselves with governments, and governments are treating AI labs the way they treat critical infrastructure.

What the deal actually covers

The MOU was signed on October 13 on the sidelines of the Singapore International Cyber Week by Anthropic co-founder Benjamin Mann and MDDI Permanent Secretary Chng Kai Fong, with Minister Josephine Teo and Anthropic’s APAC policy head Michael Sellitto witnessing. It was announced publicly at Anthropic’s Builder Summit at Capella Singapore on October 14.

The cooperation covers safety and security, model evaluation, testing standards, incident response, research into AI adoption and its economic impact, and responsible adoption across the public and private sectors. The office sits in the Raffles Place area and is hiring for sales, marketing, recruitment, and finance roles. Dale Finlay, formerly a Google Cloud executive, joined as General Manager for ASEAN.

What Singapore gets, and what it does not

The honest version: this is access, not control. Sellitto said the agreement gives Singapore no early access to Anthropic’s frontier models ahead of commercial release. Mann said the company is working to bring its most advanced models “on soil” in Singapore, but gave no dates. And the new office has no research operation planned for now; the immediate priority is sales and market development.

That modesty matters. Singapore has spent years building AI assurance credibility: Anthropic joined IMDA’s Generative AI Evaluation Sandbox in 2023, then worked with IMDA and the AI Verify Foundation on red-teaming across English, Mandarin, Malay, and Tamil. Singapore signed an applied-AI agreement with OpenAI in May 2026 and an AI safety MOU with Microsoft in June 2026. The Anthropic pact slots into a deliberate program, not a photo op.

The sharper signal came from Senior Minister K Shanmugam, who said in an interview this week that the government wants “full transparency immediately” and quicker sharing of information from AI companies as risks rise. Singapore is asking to see the models’ failures early. That is a demand no lab can meet with a press release.

The labs are following the users

Singapore is not a market Anthropic is entering. It is a market Anthropic is catching up to. According to the company’s Economic Index, Singapore ranks second out of 121 countries in Claude.ai usage relative to population, at nearly six times what its size alone would predict. More than 1,500 GovTech engineers use Claude Code. Claude is already inside services used by millions of Singaporeans every day through GovTech, Grab, DBS, GIC, and Sea Ltd.

This mirrors the Bengaluru move. There too, the pattern was the same: adoption ran ahead of the lab, and the lab followed. When your heaviest users sit in a city-state with strict reliability standards and its own AI safety institute, an office and an MOU are how you earn the right to keep serving them.

The difference with Singapore is that the government is part of the deal. In India it was an office and an announcement. In Singapore it is an office plus a formal safety pact. Governments are not just regulators standing outside the fence anymore. They are negotiating to be partners inside it.

Safety testing is becoming a two-government job

Zoom out and the week tells one story. On October 11-12, the UK government won back the right to safety-test frontier US models before deployment. On October 14, Nvidia committed $250 million in computing resources to support AI safety research in Britain. On the same day, Anthropic signed a safety MOU with Singapore. Three separate announcements, one direction: states want hands on the models before release, and labs are finding it useful to give them that.

For the labs, the calculation is not mysterious. Independent safety testing by a credible government is the fastest way to build public trust in systems that keep behaving unexpectedly. For governments, it is leverage. An MOU is not a law, but it is a relationship, and relationships turn into norms.

The open question is whether this stays voluntary. “Full transparency immediately” is not the language of a passive observer. If an incident ever forces Singapore’s hand, the MOU will be the document everyone points to, on both sides.

What this means for builders

First, if you build in Singapore or Southeast Asia, your compliance surface just grew. A safety pact between a frontier lab and your government is a preview of the questions regulators will ask you next: how you evaluate models, how you respond to incidents, how you track adoption impact. Keep your model-version logs and eval notes. You will need them.

Second, Singapore’s approach is a template. Government-by-MOU moves faster than government-by-law. Expect more of these deals in more countries, especially in markets where a lab’s user density is high. If you operate across borders, map which governments have pacts with which labs. The patchwork matters.

Third, “on soil” models are coming. Singapore wants frontier models running inside its borders, and regulated businesses there need it. If your product depends on API calls to US-based models, start designing for regional model deployment now. Sovereignty is moving from a policy debate to an architecture requirement.

And for everyone else: the frontier AI race now runs through government buildings, not just data centers. The labs that win will be the ones governments trust to test. The governments that matter will be the ones labs trust with their failures. Singapore and Anthropic just formalized that bargain.

Sources

The Straits Times, “Anthropic signs MOU with Singapore on AI safety,” October 14, 2026
Channel News Asia, “Anthropic opens Singapore office, signs AI safety agreement with government,” October 14, 2026
The Business Times, “Anthropic works to bring frontier AI models ‘on soil’ in Singapore, signs government pact,” October 14, 2026
Anthropic via X, October 14, 2026