Azeem Azhar and Paul Krugman on AI hacking, investment risk and the future of work

AI could deliver real economic gains while making the internet less secure and leaving investors with disappointing returns. In a conversation published on 31 October 2026, Azeem Azhar and Paul Krugman examine why cybersecurity, investment returns and the consequences for workers need to be assessed separately.

• Cybersecurity is entering a more dangerous transition. Azhar argues that increasingly capable, cheaper AI models can discover and combine software vulnerabilities at a pace human security teams struggle to match. He calls for faster patching and public-private coordination, particularly beyond the technology sector. His concern extends to state-backed attacks on electricity, water and financial infrastructure.
• The web may split into spaces for people and machines. Azhar expects publishers and retailers to provide more structured information for AI agents, with identity checks, delegated permissions and spending limits. People may increasingly choose protected environments rather than navigate an open web dominated by automated activity.
• Useful technology does not guarantee good investment returns. Krugman asks how much of AI's economic benefit suppliers can capture if cheaper models erode pricing power. Azhar distinguishes model developers from infrastructure providers, which can serve both proprietary and open-weight models. He argues that the key test is whether customer spending can support the financing commitments, not simply whether AI works. A prolonged period of weak returns is another possible outcome besides a spectacular crash.
• Productivity gains remain uneven. Both acknowledge that broad economic evidence is still limited. Azhar sees progress in some companies, but says the pace at which benefits spread remains uncertain. He warns that successful adoption could affect workers through weaker wage growth and fewer openings, not just visible mass layoffs.
• Compute economics depend on demand, power and local consent. Azhar argues that strong demand currently supports continued use of older chips, while warning that oversupply could sharply reduce prices. He also sees power availability becoming a more serious constraint. Both discuss the need for data-centre developers to share benefits with local communities rather than rely on construction jobs alone.
• Human judgment needs deliberate protection. Azhar closes with a concern that heavy reliance on AI could weaken independent thinking. His response includes handwriting, reading and periods of work away from computers.

This is a summary of the speakers' arguments and forecasts, not an independent verification of the technical or financial claims discussed.

Source: Azeem Azhar on the Explosion of Hacking and Other AI Issues, Paul Krugman's Substack, published 31 October 2026; conversation recorded 29 October 2026. Cover: video still of Paul Krugman and Azeem Azhar at 1:07.