ChinAI #344: AI Safety/Security Governance Report (Part II)

2022-03-07 · Source: ChinAI Newsletter · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy, Emerging Technologies & Innovation · Depth: Advanced, short

Summary

The second part of the "AI Safety/Security Governance Research Report (2025)" by CAICT highlights global developments and inconsistencies in AI governance, particularly concerning safety benchmarks. CAICT researchers acknowledge international efforts, citing techniques like lie detection for "autonomy tests" and Nanyang Technological University's INTENT-FT research for harmful intent detection, alongside third-party tools like Knostic AI and Scalekit. A key finding from Stanford University's "AI Index Report (2025)" indicates a consensus on general capability benchmarks but a lack of consistency in safety and responsible AI benchmarks among models like DeepSeek-R1, Gemini 2.5, Grok-2, OpenAI's o1, and Anthropic's Claude 3.7 Sonnet. The report also details China's AI Industry Alliance's "AI Security and Safety Commitments," signed by 22 leading companies by December 2024, which includes a voluntary disclosure initiative for safety practices. Furthermore, it explores cybersecurity governance platforms, such as the MIIT Network Security Threat Information Sharing Platform and China's National Vulnerability Database (CNNVD), as potential models for AI vulnerability reporting, noting CNNVD's practice of altering vulnerability publication dates.

Key takeaway

For CTOs and VPs of Engineering evaluating global AI safety standards, recognize the significant divergence in safety and responsible AI benchmarks between Chinese and international model developers. Your teams should prioritize establishing clear, consistent internal safety metrics and consider how voluntary industry commitments, like those from China's AI Industry Alliance, could inform your own disclosure practices. Be aware that China's cybersecurity vulnerability reporting models, if applied to AI, may introduce complexities regarding data transparency and government oversight.

Key insights

Global AI safety governance shows inconsistent benchmarking and varying approaches to vulnerability reporting.

Principles

• AI safety benchmarks lack global consistency.
• Industry alliances can drive voluntary safety disclosures.

Method

CAICT researchers analyze global AI governance by reviewing international research, third-party tools, and comparing safety benchmarks, while also exploring existing cybersecurity vulnerability reporting platforms as potential AI governance templates.

In practice

• Examine INTENT-FT for harmful intent detection.
• Investigate Knostic AI and Scalekit for AI safety.
• Monitor China's AI Industry Alliance commitments.

Topics

• AI Governance
• AI Safety Benchmarks
• Vulnerability Reporting
• China AI Policy
• AI Industry Alliances

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Ethicist, Policy Maker, Research Scientist

Related on AIssential

• ChinAI #343: AI Safety/Security Governance Research Report (CAICT 2025) — Chinese AI governance research actively integrates global findings on safety, security, and emerging risks.
• The World Has Arms Control Regimes, But AI Companies Are Not Answering to Them — AI companies are building internal arms control without external oversight, creating a critical governance gap.
• Promising Signals on AI Governance from China — China advocates for international cooperation and robust governance frameworks to mitigate AI's existential risks and ensure human control.
• ChinAI #346: Reputation Collectives - how international industry associations have helped raise China's safety standards in high-risk technologies — International industry associations improve safety in high-risk sectors by fostering "reputation collectives" through internal, confidential monitoring.
• AI incidents, audits, and the limits of benchmarks — AI safety requires systematic incident reporting and rigorous third-party evaluation to manage real-world risks.
• Links For February 2026 — AI development faces complex challenges in alignment, ethical integration, and societal impact.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by ChinAI Newsletter.