Current Implementations and Near-Misses. Currently, no frontier general model updates its weights from deployment data. All shipped personalization at the frontier—such as ChatGPT Memory, Claude memory, and Gemini Personal Context—relies on retrieval or context injection. However, the engineering for fast-cadence weight updates is already proven in specialized and smaller models. Cursor runs real-time reinforcement learning on its coding models, capable of shipping checkpoints to production
Pushed timelines back slightly to align with the expectation that labs will face multi-year safety and evaluation hurdles before sustaining high-frequency, deployment-driven weight updates on frontier models .
Current Deployments and Strict Requirements Nothing qualifying exists today at a top-five lab by revenue (OpenAI, Anthropic, Google, xAI, or Mistral). All shipped memory features at the frontier—such as ChatGPT Memory, Claude memory, and Gemini Personal Context—rely on retrieval or context injection rather than weight updates Frontier model uses daily weight updates / online learning. The resolution criteria specifically requ
Adjusted timelines outward to reflect the severe safety and poisoning risks that will likely delay any public deployment of architectures that merge untrusted customer updates into a shared base model .
As of November 14, 2026, no jurisdiction holds the authority required to resolve this question. The barrier is not a lack of regulatory appetite, but the rigid structural criteria of the mandate: the evaluations must be government-commissioned (rather than developer-paid), cover already-deployed models, and occur on a fixed recurring schedule. This combination runs directly counter to standard regulatory design. To conserve state capacity and budgets, legislatures overwhelmingly favor developer-
Set against related questions, this distribution was shifted slightly later, moving the median from early 2031 to mid-2032, to better reflect typical legislative and institutional implementation lags .
The Deployment Gate The primary bottleneck for this outcome is the deployment of shared weight-updating mechanisms at the frontier. As of November 2026, no publicly known frontier chat or reasoning model updates its weights from live deployment sessions. All shipped frontier personalization features—including ChatGPT Memory, Claude memory, and xAI's recent memory-based agents News: Research, Product & Company Updates | SpaceXSI—rely on retrieval or context injection, lea
Extended the timeline to align with the delayed expected rollout of shared-weight continual learning systems, which are a prerequisite for this specific type of cross-user contamination incident .
As of mid-November 2026, no eligible frontier AI lab (OpenAI, Anthropic, Google, xAI, or Mistral) has publicly credited a specific capability improvement in a released model to learning from customer deployment sessions. While standard periodic retraining on aggregated consumer-tier data is universal across the major labs, their public statements remain purposefully generic. For example, Anthropic's Claude Fable 5.2 system card merely notes the use of user data "included in feedback or bugs, or
Adjusted slightly to reflect the broader consensus that periodic aggregate retraining on deployment data will precede more rapid, customer-specific weight updates, aligning the median resolution with mid-2028 .
As of November 2026, no qualifying statement has been made by a large enterprise. While analysts and commentators are increasingly discussing "memory lock-in"—such as Oliver Wyman's framing of agent memory as a non-portable "book of business" How to avoid vendor lock-in with agentic AI—surveys attribute existing switching friction to agentic workflows, embedded application logic, and
Set against related questions, the median was held in late 2028 to align with the expected timeline of adapter-based personalization costs , balancing multi-homing against rising memory lock-in.
Status Quo and Required Conditions For this outcome to materialize, a top-five lab (OpenAI, Anthropic, Google, xAI, or Mistral) must make access to its absolute best model tier strictly conditional on customers granting training rights over their sessions. No lab currently does this. Existing data-for-access programs operate entirely as optional incentives. Google’s invite-only Contributor Tier offers steep discounts in exchange for training rights, but the supported models remain available
Ask a followup
Sign in to run · $20 free credit, no card · every claim cited