Payment Industry Model Can Serve as an Example for Artificial Intelligence Standards
As AI pioneers like Anthropic, Google, and OpenAI strive to establish common security standards, there are important lessons they can learn from testing infrastructures in the payments sector.
It is noted that as leading artificial intelligence companies such as Anthropic, Google, and OpenAI strive to establish an industry standards body for common security standards and independent audits, they can draw inspiration from standardization infrastructures in the payments sector, such as the FIDO Alliance and EMVCo.
The Search for Common Standards in Artificial Intelligence
Leading artificial intelligence companies such as Anthropic, Google, and OpenAI agree that powerful models need common security standards.
Achieving this partnership requires setting rules that determine who will write the tests, who will manage them, and what criteria a model must meet in order to fail.
Experience of the Payments Sector
The payments sector possesses extensive experience and infrastructure regarding standardization processes thanks to initiatives spanning many years.
Organizations like the FIDO Alliance and EMVCo are expanding their long-standing joint specification efforts to include artificial intelligence agents today.
EMVCo and Technical Requirements
EMVCo writes technical specifications for specific parts of payment systems and establishes processes to determine whether applications meet these standards.
For instance, the common core specification developed for contactless payments encompasses functions such as secure communication channels, encryption, and biometric verification.
FIDO Alliance and Delegation of Authority
Having established passwordless authentication standards, the FIDO Alliance is currently developing specifications on how users can securely delegate actions to artificial intelligence agents.
This experience offers a useful reference point for companies discussing the creation of an industry standards body to conduct testing and auditing in the artificial intelligence field.
Modeling Testing Processes
The sequence in the payments sector also serves as a guide for artificial intelligence: first, defining the requirement, specifying how it works, and establishing how the application will be tested against it.
It is of great importance for an industry organization to identify the individual capabilities it wishes to evaluate and to establish common methods to measure them.
Sectoral Coordination for the Future
While OpenAI CEO Sam Altman supported an industry testing and auditing body, Anthropic CEO Dario Amodei called for independent evaluation and greater coordination.
Meanwhile, Google DeepMind CEO Demis Hassabis suggested that frontier models should be tested for dangerous capabilities, cybersecurity risks, and abilities to bypass safety measures before being released to the market.