Safeguarding AI’s Vulnerabilities

Summary by ChatGPT-4o

Based on the FP Live session "Safeguarding AI’s Vulnerabilities" with Jared Cohen (President of Global Affairs, Goldman Sachs) and Ravi Agrawal (Editor in chief, Foreign Policy), here’s a focused breakdown of the main challenges, societal responses, and potential consequences if action is not taken:

🧨 a) Most Problematic Challenges and Risks

1. Chokepoints in AI Infrastructure

   • AI relies on a fragile global infrastructure: advanced chips, data centers, undersea cables, and critical minerals.

   • Many components are highly concentrated in a few countries or suppliers, creating single points of failure.

   • Subcomponents, not just final products, are often the cause of cascading supply disruptions.

2. Geopolitical Risk to Undersea Cables

   • 95% of global data, including financial and security data, flows through undersea cables, which are vulnerable to:

     • Espionage

     • Sabotage (e.g., recent incidents involving China and Russia)

     • Geopolitical interference in contested waters like the Red Sea or South China Sea.

3. Critical Mineral Dependency

   • AI hardware and EVs need rare earth elements and critical minerals.

   • China controls 92% of global refining and processing capacity, making others vulnerable to economic coercion or supply disruption.

4. AI Power and Energy Demands

   • AI data centers require non-intermittent base-load power.

   • The U.S. is facing energy infrastructure limits, with growing demand but limited capacity to expand or retrofit data centers.

5. Export Controls and Strategic Fragmentation

   • U.S. export controls on chips are forcing nations to choose between U.S. or Chinese AI ecosystems, undermining technological neutrality.

   • Geopolitical swing states (e.g., UAE, Saudi Arabia, India) may find their ambitions constrained by reliance on U.S.-controlled AI infrastructure components.

🛠️ b) How Society Should Address These Risks

1. Shared Visibility of Supply Chains

   • Governments and the private sector must map out the entire AI supply chain, including subcomponents and hidden chokepoints.

   • Use business intelligence (like from commodities traders) to inform national security and policy decisions.

2. Diversification of Supply and Processing

   • Accelerate development of refineries outside of China, especially in the U.S., Europe, and allied nations.

   • Reform permitting processes to allow for faster infrastructure buildout, including environmentally safer refining techniques.

3. Secure and Redundant Subsea Cable Systems

   • Reroute cables to avoid contested zones (e.g., avoid Hong Kong landings; favor Taiwan, Singapore, Philippines).

   • Improve multilateral agreements on cable repair zones and strengthen monitoring of suspicious activities.

4. Energy Policy Adjustments

   • Invest in base-load energy (e.g., natural gas, nuclear) and build next-gen data centers near reliable power sources.

   • Coordinate energy + AI infrastructure development to avoid bottlenecks.

5. Strategic Alliances with “Geopolitical Swing States”

   • Partner with nations like the UAE or Saudi Arabia to develop AI capacity in geopolitically aligned regions.

   • Support them in building sovereign tech capabilities, rather than pushing them toward China.

6. Create AI-Specific Risk Frameworks

   • Treat AI infrastructure like critical national infrastructure (CNI).

   • Develop international norms and deterrents for attacks on AI infrastructure (akin to cyber warfare protocols).

🔥 c) What Happens If We Don’t Act?

1. Supply Chain Collapse

   • One supplier going offline could paralyze the global AI ecosystem.

   • Delays, cost surges, and geopolitical standoffs could severely undermine innovation and deployment.

2. Data Blackouts or Espionage

   • Cable sabotage could result in major financial losses, national security breaches, or digital isolation of regions.

3. Dominance by Authoritarian Regimes

   • If swing states are pushed away or deprived of options, China may step in, consolidating influence over AI infrastructure and norms.

4. Loss of Democratic AI Leadership

   • The West may lose its edge in AI if infrastructure and talent are not scaled securely and efficiently.

5. Environmental & Energy Crisis

   • Without strategic energy planning, AI-related power demand could strain grids, spike emissions, or lead to energy rationing.