probability: 12%

rationale:

Status Quo and Implementation Reality Resolution requires a published specification for exporting accumulated model personalization—not merely raw chat logs—implemented by at least two of the top five AI labs by revenue (OpenAI, Anthropic, Google, xAI, Mistral). As of August 2026, there are zero qualifying implementations. While Anthropic and Google have introduced memory portability features, these are strictly one-vendor proprietary flows built on prompt-and-paste "memory handshakes" and ZIP archive uploads. These are explicitly designed as customer-acquisition tools rather than shared interoperability standards 2 sources.

Standards Pipelines Actively Bypass Memory The institutional machinery for AI standardization exists but has deliberately avoided memory portability. The Model Context Protocol (MCP), the dominant agent-interoperability standard, pushed its largest revision in July 2026 but explicitly omitted any memory or personalization-export primitive 3 sources. Similarly, the Data Transfer Initiative’s (DTI) published AI Conversation Schema covers only raw conversation logs, not accumulated personalization state, and lacks implementation by any top-five lab schemas.pub. While independent candidate schemas like Portable AI Memory (PAM) exist, they currently function as best-effort converters with no native provider support portable-ai-memory.org.

Commercial Disincentives and the "Good Enough" Baseline A severe structural asymmetry works against convergence: trailing providers want to ingest user state (import), but incumbents strongly resist facilitating user defection (export). Labs have a revealed preference for monetizing friction, leading to unilateral import pipelines rather than bilateral standards. Factoring in the strong commercial incentives for AI providers to build proprietary memory moats and enforce vendor lock-in further lowers the likelihood of a shared standard emerging. Additionally, because LLMs can effectively parse unstructured text, natural-language summaries serve as a "good enough" lowest common denominator for transferring context. This text-based workaround removes the technical pain that typically forces competitors to collaborate on a strict, machine-readable schema.

Regulatory and Competitive Forcing Functions The most viable pathway to a shared standard involves regulatory pressure, such as the EU's Digital Markets Act (DMA) mandates for continuous, real-time portability. However, regulatory forcing generally yields gatekeeper-specific compliance APIs rather than a unified cross-provider specification 2 sources. Furthermore, realistic timelines for DMA designation and subsequent compliance windows would likely push any implementation to the end of 2028 or beyond. Alternatively, challengers could theoretically adopt a shared standard as a competitive weapon against an incumbent, leveraging the speed at which consortiums like the Agentic AI Foundation can move, though there is currently no momentum for this.

Final Assessment The 12% probability reflects a demanding conjunctive requirement: a genuine published cross-provider standard and dual implementation by two top-five labs within roughly two and a half years. Given the complete absence of top-tier lab involvement in current memory-standard efforts, explicit omissions in leading protocols like MCP, and fierce commercial incentives to trap user data, a shared standard is highly unlikely to mature and see adoption by 2028-12-31.