Policy desk

Obligations: Leading AI companies such as Anthropic must grant independent assessors access to internal systems and risk teams and accept their findings.

Open-weight LLMs in recruitment show significant gender and racial biases, potentially violating compliance requirements under regulations like the EU AI Act.

Obligations: Companies operating in South Korea must comply with the government's directive to accelerate AI research and applications, while strengthening risk prevention measures for deepfakes, cyberattacks, and privacy. • Opportunities: South Korea's continued push for AI development presents opportunities for companies to participate in AI research, development, and application projects, especially in the semiconductor and hardware sectors. • Deadlines: The government plans to soon announce the next phase of its autonomous AI foundational model plan, creating a timeline for companies to prepare for potential new regulations or funding opportunities.

Obligations: AI companies must provide evidence of risk prevention measures and allow government audits, increasing compliance costs and operational scrutiny.

EU member states will form an alliance to ensure AI supply chain security, increasing compliance costs and collaboration opportunities.

OpenAI will adjust its R&D strategy to prioritize safety assessments, potentially slowing down the iteration of frontier technologies.

Obligations: Frontier AI developers must comply with consistent safety requirements under a federal framework and undergo independent audits. • Opportunities: Establishing unified safety standards and collaborative frameworks can enhance the safety of AI development and boost industry confidence. • Deadlines: The call for immediate action is made, but specific regulations and frameworks have not yet been issued.

Obligations: Companies developing AI must comply with proposed regulatory guardrails to mitigate risks.

Companies must master their proprietary tacit knowledge to build continuous learning loops and reduce reliance on single model vendors.

Obligations: AI companies must open their systems to third-party evaluators for safety audits and model alignment assessments, increasing compliance costs and operational complexity.

Companies must ensure their AI research aligns with the principles of human control and benefit for humanity, avoiding pursuits of meaningless or harmful superintelligence. • Companies should retain control over their unique knowledge and build autonomous continuous learning loops, not becoming dependent on any single model provider. • Microsoft will publish a 'Code of Conduct' for public consultation, providing clear guidance for the industry.

Obligations: AI data center operators may be subject to new environmental health regulations if the plan is adopted, requiring compliance with stricter air quality standards. • Opportunities: Companies investing in AI data centers in the US could gain from a streamlined regulatory process if the plan is implemented. • Deadlines: The plan's adoption hinges on the president's decision, with no specific timeline mentioned in the report.

Obligations : Companies need to assess the impact of this feature on their Windows 11 systems and consider manually enabling it via Group Policy or Intune to enhance security.

Obligations: Companies participating in the IIFAA working group must contribute to the development of a unified standard for AI agent identity, ensuring compliance with emerging trust frameworks.

Obligations: Companies operating AI data centers in Indonesia may be required to contribute to rising electricity tariff costs, potentially increasing operational expenses.

Companies should monitor and participate in the development of embodied-intelligence data standards to meet future regulatory requirements

Obligations: US AI developers and providers must comply with mandatory national AI safety regulations and state-level AI laws, including independent third-party assessments and event reporting. • Opportunities: Companies that proactively adopt and comply with emerging AI safety standards may gain a competitive advantage and build public trust. • Deadlines: US Congress must act before its December recess to pass federal AI safety legislation, while companies must comply with existing state-level AI regulations.

Establish unified industry standards for AI validation

Promote the large-scale application of intelligent terminal technologies

The UK bans the development of superintelligent AI that could threaten human survival.

Require companies to comply with AI governance standards

Enhance the development level of computing power facilities

Requires smart glasses and similar devices to comply with stricter regulations in public places.

Obligations: Developers of frontier AI models must submit their models for a 60-day security assessment 60 days before release, otherwise they may face removal from the market. • Opportunities: Pre-release assessments can reduce security risks for models post-release and enhance public trust. • Deadlines: The policy requires the publication of safety review guidelines by August 31st, but the deadline has passed and the guidelines have not been made public.