‘Know your agent’. Uma startup para verificar a identidade dos bots

Beltic, a startup co-founded by a Brazilian in Silicon Valley, has just raised seed funding to build a global identity and risk prevention infrastructure for the burgeoning agent economy. The $7.3 million seed round was led by Norwest Venture Partners, an American VC fund with $1 billion in assets, and included participation from Restive Ventures and Oxford Seed Fund, affiliated with Oxford University.

The capitalization is roughly a year after Beltic raised $1.5 million in a pre-seed round led by Collide Capital, joined by Latitud, Positive Ventures, and venture angels.

The company aims to address one of the major challenges of the coming years: current fraud prevention and authentication systems were designed for humans, not AI agents. When a bank tries to identify a suspicious transaction, it analyzes a series of human-related signals – device used, click speed, location, and other patterns. With AI agents, the logic is different.

Beltic's approach is to analyze these AI-driven transactions in three layers: identifying who is behind the agent making the transaction; verifying if the agent has authorization to perform that action; and ensuring the executed action matches what the user actually requested.

Mike Allan, Beltic's co-founder, explains that the first layer (identifying the agent) can be simple – the Beltic system could ask the agent to show the user's email from a merchant or bank, or it could be more robust (for large purchases) by asking the user to send a selfie or a document. The second layer is analyzed by examining the protocol used by the agent to determine if it still represents the original human intent.

The third, most complex point is ensuring that the transaction matches the original intention, as an agent may interpret a directive differently from what the user intended, potentially executing a purchase that doesn't match the request. In extreme cases, the agent may even hallucinate.

Mike, who was born in Timbó, Santa Catarina, co-founded Beltic with Isha Bhatnagar, an American who worked at Coinbase for six years on client verification. Before Beltic, Mike had founded another startup, Atar, which was acquired by Porto five years ago. Beltic is entering an still-embryonic market, with infrastructure for the agent economy beginning to be built in the last 12 months.

Companies like Visa, Mastercard, Amex, and Google have all developed their own protocols for agent interactions with retail and payment systems. The challenge is that there is no single standard, making it complex for merchants to accept payments from AI agents, as they may need to integrate with multiple protocols.

Beltic's multi-protocol approach allows virtual stores and banks to interact with various protocols without having to build their own solutions. (The Beltic product is currently in testing with some clients.) While AI is evolving rapidly, Mike doesn't believe that all transactions will be made by AI agents in the next two or three years – even the majority.

However, he expects that a significant portion of payments will be made by AI agents. E-commerce enthusiasts enjoy browsing, but finding providers, insurance, and payment processes involving high-risk automation will be done by AI agents. Many systems will be automated, with high-risk payments, decisions, and actions being executed by agents, requiring an infrastructure to ensure no errors occur.