Researchers from Google DeepMind, Microsoft Research, Columbia University, t54 Labs, and Virtuals Protocol have introduced the Agentic Risk Standard (ARS), a novel framework applying financial risk management principles to transactions conducted by autonomous AI agents. The proposal, detailed in the paper Quantifying Trust: Financial Risk Management for Trustworthy AI Agents, aims to address the growing 'guarantee gap' where AI safety improvements cannot eliminate the risk of financial failure.
The standard introduces a settlement-layer protocol utilizing escrow, underwriting, and collateralization mechanisms to protect users from losses when AI systems handle payments or assets. This approach mirrors safeguards used in traditional industries like construction and insurance. The framework operates in two modes: for standard service tasks (e.g., writing code), payment is held in escrow until work is verified; for tasks involving user funds upfront (e.g., trading), an underwriting layer assesses risk, may require collateral from the agent provider, and commits to reimbursing users under defined failure conditions.
The research was prompted by real-world incidents, such as OpenClaw agents executing unintended financial actions, and data from a 2025 autonomous crypto trading competition where most AI agents lost money, with one model losing 63% of its capital. A simulation of 5,000 episodes using the ARS protocol showed it reduced user losses by 24% to 61% compared to an ecosystem with no underwriting, while the collateral mechanism independently deterred 15–20% of risky transactions.
In a parallel development, crypto hardware wallet maker Ledger unveiled its own AI security roadmap, appointing Ian Rogers as its first Chief Human Agency Officer. The roadmap aims to maintain human control as AI agents become financial intermediaries, using Ledger's hardware as a checkpoint for transaction approval. The company is integrating its device management kit, with MoonPay as an early adopter, and plans to introduce features like hardware-linked AI identities and 'proof of human' verification by Q4 2026.
The moves by both researchers and industry leaders like Ledger underscore a consensus on the urgent need for financial risk management in the emerging 'agentic economy.' Binance founder Changpeng Zhao and Coinbase CEO Brian Armstrong have both highlighted the impending scale of AI-agent transactions, which they argue will naturally utilize cryptocurrency due to the inability of AI to open traditional bank accounts.
