15 OCT 2026 — Anthropic, the developer of the Claude AI models, opened its Singapore office on 14 October and signed a memorandum of understanding with the Ministry of Digital Development and Information (MDDI) covering AI safety, model evaluation and adoption. It is Singapore's latest agreement with a major AI company this year, after deals with OpenAI and Microsoft, and it lands a day after a senior minister said the government wants "full transparency immediately" from AI firms when something goes wrong.
Disclosure: RECATOOLS is written with the assistance of Claude, made by Anthropic, whose office opening and government agreement are the subject of this article. Readers should weigh that interest.
What the MOU covers
The agreement was signed by MDDI Permanent Secretary Chng Kai Fong and Anthropic co-founder Ben Mann on the sidelines of Singapore International Cyber Week, with Minister Josephine Teo and Anthropic's head of Asia-Pacific policy, Michael Sellitto, as witnesses. CNA reported that it was signed on Tuesday, 13 October, and the office opened the next day.
According to Anthropic's announcement, the MOU "sets out areas of cooperation", and neither side has described any binding obligations. The areas are AI safety, security and model evaluation; research on how AI is adopted and what it does to the economy; and responsible adoption in the public and private sectors. Anthropic says it plans technical exchanges with Singapore's AI safety, security and assurance institutions on model evaluation, testing standards and incident response, and will share research and data, including from its Economic Index, as Singapore develops its AI rules.
The transparency question
The timing gives the agreement its context. Senior Minister K Shanmugam said in an interview on 13 October that Singapore's key request to the major tech firms with offices here, shared with many other countries, was this: "When some things happen, can we have full transparency immediately?" CNA reported his remarks alongside the signing.
The MOU mentions incident response, but Anthropic's description does not commit it to telling the government about a problem on any particular timeline. Mann framed the value of the agreement around evaluation. "When we train a new model, nobody really knows exactly what its capabilities are," he said at a developer event, as quoted by CNA, adding that working with safety institutes such as Singapore's was "very important to understanding the risks and benefits of these new models as they're being developed".
More detail than in September
When Anthropic first announced its Singapore plans, we noted that it offered a self-measured usage statistic and no revenue, customer or headcount figures. This time it named customers. The company says more than 1,500 GovTech engineers and technical staff use Claude Code, the Cyber Security Agency is exploring how Claude could help with cyber defence, and DBS, GIC, Grab and Sea use its models in various ways. It says Grab's sales assistant, BriX, saves the company's sales teams about 40,000 hours a quarter. These are the company's own claims. One of those customers has a second interest. GIC, Singapore's sovereign wealth fund, is also among Anthropic's major financial backers, The Straits Times notes.
Headcount remains undisclosed. "We're not sharing numbers on this," Dale Finlay, Anthropic's general manager for ASEAN, told CNA. CNA counted 12 job openings listed for the Raffles Place office.
Singapore's growing list of AI partners
Singapore has been signing with each of the large AI developers in turn. CNA notes the OpenAI deal in May and an IMDA–Microsoft agreement on AI safety and security in June. In the same week as the Anthropic signing, GovTech announced work with Google DeepMind on an on-device scam detector. Anthropic's relationship dates back to 2023, when it joined a generative AI evaluation sandbox run by IMDA and the AI Verify Foundation, followed by a red-teaming project in English, Mandarin, Malay and Tamil.
What would show it works
Singapore is keeping close to every frontier developer at once through agreements with broad public descriptions. This one will matter if the technical exchanges and incident-response cooperation produce something the public can see, such as joint evaluation results or a clear process for disclosing serious failures.
View author profile → · Editorial policy
