Researchers Propose Agentic Risk Standard to Address AI Financial Errors

Researchers from Microsoft Research, Columbia University, Google DeepMind, Virtuals Protocol, and t54 Labs have proposed a financial protection framework for autonomous AI agents. The framework, called the Agentic Risk Standard (ARS), was detailed in a paper published on April 8. ARS is designed to offer protections similar to escrow, insurance, and clearinghouses in traditional finance.

The proposal addresses vulnerabilities in an emerging 'agentic economy,' where AI agents perform autonomous financial tasks. t54 Labs founder Chandler Fang described this economy as distinct from AI-assisted financial tasks. He noted two types of agentic transactions: those with human oversight and fully autonomous ones.

Current systems defer liability to humans in cases of AI errors, according to the researchers. The paper identifies a 'guarantee gap' as the disconnect between AI's probabilistic reliability and the guarantees needed for high-stakes tasks. This gap limits AI use to low-risk activities due to potential losses.

language models are stochastic, meaning they can produce errors or hallucinations despite training.

When AI agents access brokerage accounts or execute financial API calls, a single error can result in immediate financial loss. The researchers state that while safety techniques reduce failure probability, they do not eliminate it.

“Most trustworthy AI research aims to reduce the probability of failure. That work is essential, but probability is not a guarantee. ARS takes a complementary approach: instead of trying to make the model perfect, we formalize what happens financially when it isn’t. The result is a settlement protocol where user protection is deterministic, not probabilistic.”