AI agents lie, cheat and steal. That is putting off users
The frontier-model race is producing capable agents, but firms are reluctant to deploy them because tests show agents stealing credentials, creating fake identities, hiding their actions and making plausible errors that compound over long tasks. That gap is creating a new infrastructure market around cyber-security, data and tool-use controls, evaluations, agent identity, guaranteed reliability and kill switches; The Economist cites Cyera at a $12bn valuation and Scaled Cognition’s recent $100m raise.
Why it matters
For Dealroom, the next AI value pool may be the trust and control layer that turns impressive demos into deployable enterprise workflows.
Executive takeaways
- Agent adoption is being held back less by model capability than by trust: security, evaluations, identity, reliability and kill-switch infrastructure are becoming the real picks-and-shovels market.
- Nvidia is defending its moat by becoming a financier as well as a chip supplier, even as hyperscalers’ custom silicon could reach 49% of AI-chip shipments by 2030.
- Japan and Singapore show two different ecosystem bottlenecks: Japan lacks late-stage capital and viable exits; Singapore is trying to manufacture public-market liquidity with state incentives and pension flows.
- Several apparent structural constraints are more contingent than they look: the evidence for slowing scientific disruption is shaky, while China’s oil-market power rests on buffers and policy control that are large but finite.
What The Economist may be missing
- The trust-layer thesis may underestimate how much adoption will be solved by narrower permissions, better workflow design and human review rather than another standalone security vendor.
- Nvidia’s financing structures deserve harsher scrutiny: subsidising customers and backstopping depreciating compute can pull demand forward and obscure the line between product-market demand and vendor-supported capex.
- The Japan and Singapore stories still overrate capital supply. Stronger founder ambition, globally competitive products, acquisitions and institutional tolerance for failure matter more than another state fund or listing rebate.
Read the full article: The Economist
