The Risky Boat: Perils of Closed-Tech Dependence
Living by the sea in Whitstable, I’m often reminded of the long-standing traditions of maritime code. Out there, the rules are unforgiving and straightforward: a captain’s first duty is to their own vessel and crew.
If another ship is in distress, you render assistance only if it doesn’t endanger your own.
It’s not selfishness. It’s survival. And it’s written into the code because the ocean doesn’t care about good intentions.
We’re seeing the same dynamic that’s always governed Big Tech.
The Wave That Hit: GPT-4o Retired Overnight
When OpenAI retired GPT-4o without warning last week (August 2025), the backlash was immediate and fiery.
- Developers saw products break overnight. Some couldn’t serve customers or deliver campaigns, losing both time and revenue.
- Many users flooded Reddit and X to complain that their workflows were crippled.
- Others spoke of emotional attachment: they preferred 4o’s tone and personality, describing GPT-5 as “colder” and “less human.”
The outcry was so fierce that CEO Sam Altman admitted it was a mistake, and within 24 hours, OpenAI restored access for paid users – but disruption had already been caused.
This wasn’t just about technical performance. It showed that people (and businesses) had built deep dependencies on specific models. One user even said 4o had been more encouraging than their parents!
That’s a different level of attachment than most technologies inspire. And when a model is pulled suddenly, the shockwaves go far beyond lost features; they hit trust, stability, and the very identity of products and services.
Every workflow we build on someone else’s platform carries hidden risk, whether we realise it or not.
Captains Will Always Save Themselves First
OpenAI didn’t act out of malice. They steered their ship where they thought it needed to go. But everyone tied to their deck was forced to follow.
This is the essence of the maritime code: a captain preserves their own ship first. And in AI, the captains are the model providers and hold an immense level of power.
If OpenAI, Anthropic, or Google decide to retire a model, change pricing, or pivot strategy, they’ll do it. Their obligation is to their own trajectory, not to your business continuity.
The GPT-4o episode was just one incident. But it revealed the bigger risk: entire ecosystems are now lashed to a handful of closed-tech providers.
Building Your Lifeboats
These are systemic (and growing) risks. You can’t rely on anyone else to save you. Plan your lifeboats before you set sail:
- Diversify: don’t rely on a single provider; design modular architectures so models can be swapped when one fails.
- Own your data: don’t let all your business knowledge live inside someone else’s system.
- Adopt open-weight models: although they may not match frontier performance today, they’ll run when external APIs are unavailable.
- Design fallback workflows: ensure critical processes can survive for days or weeks without any AI assistance.
- Document and cross-train: capture how your AI systems work and ensure multiple team members understand them. Vendor dependencies shouldn’t become people dependencies.
- Plan for total outage: what happens if models are restricted by governments, compromised by security breaches, or simply unavailable? Your business should have answers before the lights go out.
The goal isn’t to avoid AI (and other closed tech), it’s to use it without becoming helpless when it’s taken away. Because in the maritime code, when the weather turns, you’re on your own.
We learned this lesson during our ISO27001 transition. It opened our eyes to hidden risks we’d been ignoring. If you can’t afford to lose your business, you can’t afford to ignore these dependencies.
If you’re looking at your tech stack and wondering where the hidden dependencies are, that’s precisely the kind of challenge we help businesses untangle.