Policy desk

On-device LLMs offered in China must complete CAC filing before launch; devices without updated filings (e.g. iPhone 15/15 Plus) do not carry the AI features

Clarifies AI's position as a teaching auxiliary tool rather than a teacher replacement

Obligation: institutions deploying AI-automated payment systems must implement risk management requirements for high-risk transactions above VND 10 million (individuals) and VND 50 million (organisations).

AI-generated marketing content using real people's likeness or voice is a primary enforcement target, risking account bans and content removal

ADAS providers must enhance detection capabilities for stationary or slow-moving vehicles

Enhanced public safety: Real-time detection of violence and vandalism enables faster police response

Obligations: The FAA is obligated to ensure the AI system (like SMART) it deploys for air traffic management meets safety standards and provides reliable routing advice to controllers, while the system itself does not directly control planes. • Opportunities: The AI system can help optimize air traffic flow, reduce congestion, and improve airport operational efficiency. • Deadlines: The system was deployed in the airspace where the American Airlines/Black Hawk collision that killed 67 people in January 2025 occurred, requiring assurance of its safety and effectiveness.

Obligations: Companies deploying Meta smart glasses must comply with Meta's privacy guidelines, implementing the indicator light system and applying software updates to prevent vulnerabilities.

Must equip AI models with 'kill switch' functionality and establish independent auditors; - Must collaborate with the expert panel to provide recommendations within two months; - Must comply with California's AI safety regulations, facing compliance risks otherwise.

Obligations: Companies must adopt the Lanyun Yuan Sheng Dai tool for efficient and accurate pre-listing compliance checks, ensuring all key fields across documents are consistent and compliant with regulations like the Advertising Law.

Obligations: OpenAI is required to track, investigate, and publicly disclose instances of model misalignment according to the new framework, which sets criteria and timelines for disclosure.

Obligations: Companies must exercise caution when using AI tools, not blindly trust or grant permissions to unknown AI services. They must set clear behavioral boundaries and strict permissions for AI agents, not casually open internet access or content editing rights.

Obligations: AI companies must follow OpenAI's framework for disclosing model misalignment issues to the public, especially for significant or safety-related findings.

Obligations: Companies must establish internal reporting mechanisms for AI model misalignment, ensuring timely investigation and disclosure. • Opportunities: Early detection and public disclosure of misalignment can enhance model safety and transparency, potentially leading to improved public trust. • Deadlines: Companies need to implement the reporting framework within a reasonable timeframe to comply with OpenAI's new standards.

The government will support the promotion of new energy vehicles and green smart home appliances in rural areas, providing market opportunities for related enterprises.

Obligations: Companies engaged in geological exploration are encouraged to adopt this system for enhanced efficiency and accuracy. • Opportunities: Early adopters can gain a competitive advantage in geological surveying and resource discovery. • Deadlines: No specific deadlines mentioned, but companies should plan for integration and training.

Companies should pay attention to investment and collaboration opportunities in the biotechnology and life sciences field by the Hong Kong government.

Provides service for unscrambling fraud methods and risk judgment

For companies: Large-scale application of eSIM technology, development of offline communication technology, standardized process for new business filing.

Offers AI intelligent question and answer to help users identify scam risks.

Companies may consider the use of copyrighted text for AI training as fair use.

Companies can train AI without excessive concern for copyright infringement risks.

Legalizes the use of copyrighted materials for AI training, providing new opportunities for the AI industry.

For companies: Ensure the accuracy of AI medical record tools, correct errors in a timely manner, and enhance patient trust in AI technology.

Financial institutions must build human-approval mechanisms for high-risk AI uses, defining scenarios where agents need human sign-off

Internet-exposed or outdated PLCs are high-risk targets; operators must immediately audit and harden external exposure

Food delivery businesses must maintain kitchen hygiene standards to avoid violation detection