27 March 2026¶
Weekly Summary Letter
2026/03/27 Monthly Founder's Letter
Hi NUMBERS,
This is Tammy, closing out March with you. This month the AI industry learned a lesson it should have seen coming — and it lands squarely in the territory Numbers Protocol was built for.
By now you may have seen the story: a Meta AI agent passed every identity check it faced, then accessed internal data it had no clearance to touch. The system knew who the agent was. It never asked what the agent was allowed to do. Security researchers call this the "confused deputy" problem — a trusted system misusing its own authority. But there is a deeper issue the industry is only beginning to confront: when AI agents act autonomously, every step they take becomes a process that needs to be auditable. Not just the final output, but the chain of decisions, tool calls, and data accesses along the way.
This is exactly the debate playing out between NVIDIA's NemoClaw and OpenClaw. At GTC, Jensen Huang pitched NemoClaw as the enterprise answer: sandboxed execution, declarative policy, built-in audit logs. OpenClaw's community counters with transparency — open transcripts, portable memory files, and the argument that auditability requires open-source visibility, not just corporate guardrails. Both sides agree on the premise: an AI agent whose process cannot be traced is an AI agent that cannot be trusted. That is provenance — not just for media, but for machine behavior itself. And it is the same principle that has driven Numbers Protocol from day one.
On the product front, we secured a new C2PA certificate this month, and you will see it reflected in ProofSnap very soon. Some of you have asked about the difference between ProofSnap and Capture App — the distinction matters. ProofSnap is built for end-to-end provenance: every pixel, every metadata field, cryptographically sealed from the moment of capture. It is the tool you reach for when evidence preservation is non-negotiable — journalists, legal professionals, insurance adjusters. Capture App, on the other hand, is designed for creators who need to prove ownership of their work and register it on-chain. Different use cases, same trust layer underneath.
The team also published a deep dive into what lives inside a NID this month. When an AI agent queries a registered asset, a single API call returns the full provenance record — creator identity, on-chain address, cryptographic proof hash, and chain of custody. A verified agent acting on unverified content is still a broken trust chain. The NID closes that gap. Meanwhile, our "Guess the AI" campaign moved into Phase 2: registration is closed, the guessing game is live on Numbers Mainnet, and the community is sorting through image pairs to separate real from generated. Ten thousand NUM on the line — have you tried yet?
On the partnership side, we announced a collaboration with ARPA Network, combining provenance with cryptographic proof — two protocols united by the conviction that unverifiable claims are not acceptable. And a media partner has now integrated our pipeline end-to-end, preserving C2PA credentials through every resize, crop, and format conversion from capture to publication. The rails we have been building are carrying real traffic.
April will bring the new C2PA certificate live in ProofSnap, continued Season 2 community activities, and more work on the intersection of agent auditability and content provenance. The machine economy does not slow down, and neither do we.
As usual at weekends, the team may respond a little more slowly while taking some rest. Our moderators will keep supporting you, and we'll be well prepared for the new week. One small request, keep amplifying Numbers on X and beyond. Your shares make a real difference.
Best regards,
Tammy