Strengthening Cybersecurity Amid Agentic Commerce
Matthew Driver, the executive vice-president for services in the Asia Pacific at Mastercard, emphasizes the need for robust cybersecurity measures in response to the complex software environment and the emergence of agentic commerce, where autonomous AI agents independently manage online transactions for customers.
Control in Agentic Commerce
- Agentic commerce involves AI acting on behalf of customers, requiring transitions from KYC (Know Your Customer) and KYB (Know Your Business) to KYA (Know Your Agent).
- Essential components include:
- Ensuring customer consent and instructions to the agent.
- Verifying the agent and its data sources.
- Tokenizing instructions to establish provenance.
- Providing recourse for errors made by agents or merchants.
Major Use Cases in Agentic Ecosystems
- Key focus on convenience, such as AI summaries and machine discoverability.
- Enhancements include:
- Application programming interfaces for inventory and pricing.
- Generative engine optimization for discoverability.
- Tokenization, passkeys, and biometrics for seamless checkout.
- Examples include price tracking and automated ticket purchasing for events.
Addressing AI Model Vulnerabilities
- Advancements in AI, like Claude Mythos by Anthropic, present both opportunities and threats.
- Mastercard maintains a secure payment network and collaborates with AI companies to manage threat vectors.
- Public-private partnerships and threat intelligence are crucial for mitigating cybersecurity risks.
Cybersecurity Spending and AI Utilization
- Speculation exists about increased cybersecurity spending due to AI advancements.
- Focus is on efficient AI usage and return on investment.
- There's an anticipation of increased small language model usage.
Ensuring Robust AI Agents in Transactions
- Monitoring of 18 to 19 variables, including agent behavior and model drift.
- Establishing agent registries and network governance.
- Setting standards for performance and imposing penalties for violations.
