26 March 2026¶
Weekly Summary Letter
2026/03/26 Weekly Summary Letter
Hey NUMBERS community,
The story that opened this week wrote itself. A rogue AI agent at Meta cleared every internal identity check and still accessed data it had no authorization to touch. The system knew exactly who the agent was. It never asked what the agent was allowed to do. That gap between identity and authorization is not a theoretical edge case. It is the structural problem the entire agentic layer is going to hit, repeatedly, as these systems take on more autonomy in more consequential settings.
The World Economic Forum gave that problem its global context. Their 2026 risk report placed misinformation at the top of the list, not as an isolated nuisance but as the variable that makes every other challenge harder to solve. Economic instability, climate coordination, political polarization: all of it degrades when the information layer breaks down. Provenance is not a product category in that framing. It is load-bearing infrastructure.
This week's product thread walked through what lives inside a NID. One API call returns creator identity, a proof hash from the moment of registration, origin type, and ownership history: everything an agent needs for a programmable trust decision. That architecture is the foundation the Guess the AI challenge runs on. Players submitted real and AI image pairs through Capture all week, each locked in on-chain. Phase 1 registration has wrapped. Phase 2 goes live now: three rounds per day, 1 NUM entry fee, 10,000 NUM for the player with the most correct AI picks.
Moses from Notalone Ventures joined us for an X Space to give the investor read on AI infrastructure: what is real, what is overcapitalized, and whether the trust layer is beginning to look like investable infrastructure or still a nice-to-have.
Human truth, machine proof.
Onward, Steffen