Aramco Ventures has co-led a $20 million seed round in Twin1 AI along with Bessemer Venture Partners and Tribeca Venture Partners. Exiting stealth, the funding will help the AI startup hire in San Mateo, where it is based, and London. The funding will also support go-to-market spend and further work on its core technology.
Founded in 2025 by Lewis Z.Liu, Tom Cahn, Jonathan Budd, and Huiting Liu, Twin1 says the twins are built from a user’s emails, meetings, documents and workplace systems, and run inside Slack, Microsoft Teams, Outlook, Gmail, Google Drive and SharePoint, answering questions and taking approved actions on a person’s behalf.
The founding team is the same as the one behind Eigen Technologies, and several investors who backed that earlier company returned for this round.
Aramco’s investment arm points to that governance work as its reason for backing the company. “Serving the world’s largest and most highly regulated enterprises requires far more than checking the standard security and compliance boxes,” said Daniel Carter, Managing Director of Aramco Ventures US, in news repords. He added that the founders combine an understanding of how such organisations operate with the experience needed to make AI work inside them.
The platform also exposes an enterprise MCP server, letting other AI agents and internal tools draw on a twin’s context, and applies six layers of rules-based and AI-based permission controls over what any twin can share. The company offers SaaS, single-tenant and private-cloud deployment, and describes the result as sovereign AI that reduces a customer’s dependence on any single model or infrastructure provider.
Twin1 says it has been running with partners in legal, financial services and energy for more than a year, and names Linklaters, Orrick, Dechert, Customers Bank and Aegis Energy among its customers. Orrick also took a strategic investment position in the seed round, and the firm’s Chief Innovation Officer, Wendy Butler Curtis, is quoted in the launch announcement describing the platform as a way to mine the firm’s collective data and improve client advice.
Those customers are also the source of the company’s automation claim. Twin1 reports that users say the platform automates 30 to 50 per cent of the communications work done by knowledge workers. No baseline, measurement window or methodology accompanies the figure.
Chief executive Lewis Liu said the point of the technology is to amplify individual expertise rather than average it out, arguing that in a knowledge organisation the human is the atomic unit of knowledge and that AI should preserve a person’s judgement and voice.
The bet in this round is not on another agent or another model, but on the layer that decides what an agent is allowed to see, and three co-lead investors paying $20 million at seed stage for that layer tells you the category is now fundable on its own.
It also puts a clock on Microsoft, Google and the agent vendors, who will have to answer the permissions question natively before a startup owns the answer.
The buying landscape has shifted. Purchasers now focus on strict data control rather than technical capability alone, bringing general counsel into purchasing decisions alongside chief information officers.
This change is particularly vital across the region, where banks, energy companies, and government entities have delayed adopting automated software over privacy and security concerns. By combining in-country data hosting with individual user permissions, companies give procurement teams concrete controls they can test directly against SDAIA and PDPL rules, moving the process from abstract policy debates to practical verification.
