This low probability reflects the stark absence of any binding threshold today, the short ~25-month window until the end of 2028, and a clear regulatory preference for developer-defined safety criteria. Currently, no government holds recurring, independent re-testing power over a deployed frontier model. Furthermore, because no frontier lab yet updates a flagship model's weights from live user sessions, there is little immediate pressure on lawmakers to regulate continuous or online learning.
Set against related questions, this forecast held steady at 7%, as its timeline remains consistent with the broader consensus that strict, government-run capability testing regimes are highly unlikely to take effect before the end of 2028.
Current Safety Frameworks As of mid-November 2026, none of the three specified safety frameworks contains a re-evaluation trigger for continuous, versionless weight updates. Anthropic's Responsible Scaling Policy (still v3.4) relies on periodic Risk Reports and assessments keyed to compute multiples or accumulated post-training anthropic.com. Google DeepMind's Frontier Safety Framework v3.1 uses "substantial modification" and "Material Capability Change" tests that compare discrete versio
Resolving this requires computing chance-adjusted probabilistic agreement on model errors (CAPA) across the five highest-scoring organizations' flagship models. However, this measurement has never actually been run on frontier models; the published CAPA pipeline remains frozen at early-2025 open-weight models 2 sources. Since there is no recorded baseline from August 2026, any assessment depends on what a comparable retrospective evaluation would show. Consequently, the probability relies
Current State and Feasibility As of late 2026, no frontier AI lab (OpenAI, Anthropic, Google, xAI, or Mistral) serves customer-learned weight variants of a flagship model. All deployed personalization relies on context injection, retrieval, or memory (such as ChatGPT Memory or Gemini Personal Context), which do not alter model weights. The binding constraint on serving thousands of personalized variants is not technical feasibility: multi-adapter serving techniques like S-LoRA have already d
Decreased slightly to 8% to maintain consistency with the expected persistence of context-based personalization and the retreat from massive per-customer weight variants .
As of November 2026, none of the five frontier AI labs—OpenAI, Anthropic, Google, xAI, or Mistral—include explicit language in their standard enterprise terms making accumulated model personalization non-portable on termination. Current contracts either remain silent on the matter, rely on generic intellectual property reservations, or simply stipulate the deletion of customer data. Google’s terms even preserve customer access to tuned models after termination where applicable [Service Specific
Set against related questions, this was lowered from 9% to 7% because the expected lack of standard export protocols means providers can rely on technical friction rather than restrictive contractual language.
Current State and Resolution Requirements Resolving this question requires both a published, cross-provider specification for exporting accumulated model personalization (memories, preferences, or learned weights) and implementation by at least two of the top five AI labs (Anthropic, Google, Mistral, OpenAI, xAI). As of mid-November 2026, no such standard exists. Existing interoperability tools are strictly vendor-specific import features designed for customer acquisition. For instance, Anth
Set against related questions, this estimate was held stable, reflecting the view that leaders will rely on technical friction to constrain switching rather than adopt shared export standards.
Ask a followup
Sign in to run · $20 free credit, no card · every claim cited