AI model retirement, trade bans expose fragility of centralized artificial intelligence
When major artificial intelligence platforms abruptly retire popular models or alter user access terms, developers and professionals have often watched months of tailored context disappear overnight. The risks are exacerbated when governments view these technologies as threats to national security, and make sudden policy decisions at the expense of existing users. The United States swiftly banning Anthropic from offering their top-shelf AI models, Mythos 5 and Fable 5, and sudden policy flip-flopping, also shows why relying on centralized systems without a back-up plan is inherently risky, industry analysts say.
"People who have incorporated AI into their work and personal lives already know that their AI memory is an extraordinarily valuable asset," said Brian Hankey, founder of European AI firm Penfield. "Every conversation, every correction, every connection between ideas. That’s yours. Or at least it should be."
Penfield makes it possible for people to be more resilient to Big Tech while still making use of their AI advances, by creating an autonomous system that "remembers" your use of AI, even when the big platform models constantly forget. "The platforms will always act in their own interest because that’s what platforms do. Expecting them to protect your data is like expecting the landlord to care about your furniture," Hankey says.
AI users need independence, more ownership
Most internet users have already experienced the shock of a platform changing the rules overnight. With AI assistants, the stakes are higher: it's not just a follower count or a content feed that disappears, it's months of accumulated working knowledge, Hankey argues. When a centralized platform retires a model, updates its terms of service, or locks an account, all of it can vanish instantly. For employers, this also means a significant waste of time and resources.
Currently, just over half of Dutch workers think AI will increase productivity, according to Statistics Netherlands. Research center TNO said this is one issue employers would be wise to examine moving forward. A limited study by the Dutch center pointed out that when AI takes over certain tasks, it may make an employees' day-to-day life too scattered, or potentially more monotonous, depending on how the extra time is utilized.
For many professionals who rely on artificial intelligence agents, few tasks are more tedious than repeatedly typing the same foundational guidance and project background into a chat window, or trying to recreate a set of prompts to retain their AI agents’ personality. Hankey says the lack of long-term and “inevitable enshittification” by the top platforms means retention is the primary bottleneck for AI adoption, limiting the practical utility of otherwise sophisticated digital assistants.
Seamlessly switching between different AI models without sacrificing users’ history of corrections and working styles is one way to quickly increase productivity without having to repeat the same rote prompting tasks over and over again.
The Penfield solution to enshittification, or “platform decay”
The technology operates as an independent software layer, meaning it functions across various ecosystems including desktop interfaces, mobile devices, and virtual private servers. Users can establish a secure link to their preferred assistants through native connectors in about 30 seconds, bypassing the complex configurations usually required by developer-targeted tools. This approach ensures that a user's intellectual property remains exportable and completely portable regardless of external corporate decisions.
“We designed Penfield with users’ needs in mind, so when companies make their AI models worse by making them less natural, or try to push more expensive products, Penfield subscribers will barely notice,” Hankey said. “Our customers can be AI platform agnostic, easily switching from one model to another, from one provider to another, and still take their AI agents and past use history with them when they make a change.”
Penfield was created as a turn-key solution for normal users, with more advanced options for developers. It is compatible with Anthropic’s Claude, ChatGPT from OpenAI, Microsoft CoPilot, or even Google Gemini. “Really, any service that uses the Model Context Protocol standard, that is to say any service actually worth using,” said Hankey.
To address these vulnerabilities, the platform provides persistent memory management for artificial intelligence assistants. Having launched its services two months ago, the company has just started signing up new customers who require complete data independence over their digital assets.
To encourage early adoption, Penfield is running a promotional campaign that offers a one-month free trial, followed by 50% off the first year of service. The limited-time offer brings the standard tier price down to 10 dollars per month, and accommodates 5,000 memories and 25 documents, while the premium tier costs 20 dollars per month to support 20,000 memories and 100 documents. Both options include knowledge graph access, vector search, and API access, and all paid plans come with the free trial period.
As a company headquartered in the European Union, Penfield pledges to follow EU rules about privacy and data sovereignty, not only for chat history but also for cloud-based document storage. The company is also planning enterprise subscriptions and solutions, like offline use for organizations in sectors that have specific compliance requirements.