AI Governance Standards Gain Momentum
The AI Boundary Declaration Protocol (AIBDP) has been submitted as an Internet-Draft to the IETF, proposing a technical standard for AI system boundaries. This comes as India publishes its own AI governance guidelines and the ISO/IEC 42001 standard for AI Management Systems gains traction, signaling a global push for standardized AI governance.
- The AI Boundary Declaration Protocol (AIBDP) is an individual submission to the Internet Engineering Task Force (IETF), not yet an official standard; its adoption is voluntary and aims to create a machine-readable way for websites to grant or deny permissions for AI training and use, similar to how `robots.txt` guides search engines. - India's framework avoids creating new, standalone AI legislation, instead leveraging existing laws like the Digital Personal Data Protection (DPDP) Act of 2023 and establishing new bodies like an AI Governance Group and an AI Safety Institute. - ISO/IEC 42001, published in December 2023, is the world's first certifiable international standard for AI Management Systems, providing a formal framework for risk management that aligns with the principles of major regulations like the EU AI Act. - Blockchain technology is being explored as a foundational layer for AI governance, offering an immutable and transparent ledger to track data provenance, model changes, and compliance, which can help in auditing AI systems for bias or manipulation. - The push for AI governance is creating opportunities for decentralized infrastructure, where AI-focused cryptocurrencies are used to power networks that allow users to get paid for renting out spare computing power or for the use of their data in training models. - Smart contracts could be used to automate AI compliance and governance, for instance, by programmatically enforcing fairness standards or halting a model's deployment if certain ethical or regulatory lines are crossed. - The IETF's work in this area, through its AI Preferences (AIPREF) working group, is focused on creating a more granular vocabulary than `robots.txt`, allowing publishers to specify different permissions for different AI activities like indexing versus model training. - Regulations like the EU AI Act mandate that high-risk AI systems maintain automatically generated logs and technical documentation for up to 10 years to ensure traceability, a requirement that aligns with the strengths of blockchain for secure, tamper-proof record-keeping.
