Anthropic has located a hidden subspace inside Claude where concepts are processed before being output. At the same time, OpenAI is accelerating its super app strategy, combining chat, voice assistant, productivity tools, and payments. Two stories that seem unrelated – but both hit the same nerve: control over AI depends on transparency and vendor lock-in.
Anthropic's discovery is technical, but the implications are political. Knowing how Claude 'thinks' means you can audit for alignment, bias, or hallucination. For Italian SMEs that are starting to embed AI into customer support, inventory, or accounting, this is the line between a reliable tool and a black box. One mistake from a chatbot that handles bookings or invoices can cost real money. No business should run an AI it cannot understand.
On the other hand, OpenAI's super app (modeled after WeChat) promises convenience but creates a closed ecosystem. For an Italian company, handing over chat, email, payments, and AI to a single US provider means losing control over data – exposed to the Cloud Act, unilateral price hikes, and policy changes. We have seen API prices triple in a year. An entrepreneur building on top of a foreign super app risks being held hostage by a single vendor.
Sponsored Protocol
Our position: transparency and digital sovereignty are not slogans, they are engineering decisions
We, at Meteora Web, have always favored open, controllable stacks. Not out of nostalgia for hand-coded solutions, but because we know what it means for a client to depend on a system they cannot influence. Explainable AI is not an academic luxury: it's a requirement for GDPR compliance, for auditing, and for proving to your own customers that your software is fair. And the super app? It's the opposite: data hostage, zero interoperability, rising costs. Cybersecurity in Italian SMEs is already weak; adding a monolithic proprietary layer is like putting a cheap lock on a reinforced door.
Sponsored Protocol
Europe has the AI Act, but fines alone won't create awareness. Those who build for clients have a duty to explain the risks. We do it daily: when a client asks us to integrate ChatGPT, we start with the right questions – what data does it need, where does it go, who sees it, what does it really cost. That's not fluff: it's accounting.
What to do: don't wait for regulators to decide for you. If you're evaluating an AI tool, ask your vendor – and yourself – how it works internally. Demand explanations. Prefer open-weight models or those with independent audits. And if someone pitches a super app to 'simplify' everything, read the contract with your accountant, not your marketer.
