August 8, 2025
5 min read
Lu-Hai Liang
A world of powerful AI Agents needs new identity framework
The rapid advancement of AI, particularly with tools like OpenAI's ChatGPT, is creating significant challenges for the digital landscape, especially concerning AI agents. Peter Horadan, CEO of Vouched, highlighted these issues during a recent Dock Labs webinar, emphasizing the need for robust identity frameworks for these increasingly sophisticated AI assistants. AI agents, akin to personal assistants, can now perform actions on our behalf, such as booking flights. This capability, while convenient, raises cybersecurity concerns as users might be prompted to share credentials, effectively granting session keys to AI agents. This practice, according to Horadan, is "terrible training" that conditions users into sharing sensitive information unsafely. Currently, AI agents often use methods like screen scraping, impersonating users to log in and execute tasks. While some frameworks like Anthropic's Model Context Protocol (MCP) offer more controlled interaction, they lack crucial identity management features.Key Challenges for AI Agent Identity
- Distinct Identification of Agents: It's vital to clearly distinguish between a human user and an AI agent when actions are performed. This requires mechanisms for distributed authentication and role-based delegation to precisely track the authority granted to an agent.
- Reputation Tracking: Similar to how email systems grappled with spam and phishing, the AI agent ecosystem will likely see both beneficial and malicious actors. A reputation system, like a "Yelp for AI agents," is needed to monitor behavior and identify potentially harmful agents.
- Legal and Contractual Considerations: Traditional terms and conditions assume human consent. When an AI agent consents on behalf of a user, the legal validity becomes questionable. Clear pre-negotiated frameworks or prompts for explicit human confirmation are necessary.
- Durable, scoped authorizations: Agents would present session keys for permitted actions.
- Clear agent credential identification: Agent identity would be separate from user identity.
- Reporting mechanism: Service providers could submit feedback to an impartial rating authority. This integrated approach, termed MCPI, forms a new identity layer for MCP, with demonstrations showing its compatibility with existing IAM and CIAM systems, as well as its integration with mobile driver's licenses (mDLs), European Digital Identity (EUDI), and verifiable credentials.
- AI Agents: Are They Broken? Can GPT-5 Fix Them?
- AI Crypto Trading Tools Reshape Market Strategies in 2025
- Understanding the Importance of Cryptocurrency Compliance
Proposed Solutions
Vouched has put forth a "Know Your Agent" framework and an Identity Extension for MCP. This proposal, inspired by OAuth 2.0, aims to enable:Digital Identity Rights Framework (DIRF)
A separate research paper, the "Digital Identity Rights Framework" (DIRF), has been proposed by researchers from institutions including Nokia, Deloitte, and J.P. Morgan. This framework is designed to protect behavioral, biometric, and personality-based digital likeness attributes in the face of widespread generative AI. DIRF addresses digital identity protection and clone governance in agentic AI systems, defining 63 enforceable identity-centric controls across nine domains, categorized as legal, technical, or hybrid. These domains cover aspects like identity consent, model training governance, traceability, memory drift, and monetization, aiming to safeguard individuals from unauthorized use of their digital identity. Notably, DIRF has also shown potential to enhance LLM performance by improving prompt reliability and execution stability. An implementation roadmap is provided, and DIRF is compatible with security frameworks like NIST AI RMF and OWASP LLM Top 10.Source: Originally published at BiometricUpdate.com on August 8, 2025.