Cybersecurity risk: ChatGPT 5.5's code filters are a disaster, say paying users

ChatGPT 5.5 blocks coding: Censorship or protection?

For the past two weeks, thousands of ChatGPT Pro subscribers (€200/month) have been met with an unexpected message when trying to generate code: "cybersecurity risk". Version 5.5 of OpenAI's model has introduced filters that block everything from improving an ERP to building a simple website, something that did not happen with version 5.4 or with competitors like Claude or Grok. According to some analyses, the measure is meant to shield OpenAI from potential legal liability, but the developer community calls it excessive and counterproductive.

The generative AI containment wall

The restriction is not arbitrary. Advanced language models can generate code that, without supervision, could be exploited. However, the decision to block any programming request, even the most harmless ones, has sparked a wave of criticism. An analysis of the debate shows two opposing camps: those who believe the measure is necessary to prevent AI from becoming an exploit factory, and those who see it as a "lobotomization" that reduces the product's usefulness. A telling detail: OpenAI offers a registration as a "computer security company" that requires excessive documentation, but even that does not guarantee access.

Alternatives and the debate on the future of programmers

The exodus to other tools is palpable. Claude Opus 4.7 has become the favorite of many, although it requires very precise instructions to avoid monolithic code. Grok has proven capable of generating complex applications, such as a 5-axis CNC controller in Python, but its reliability depends on the operator. At the heart of the debate is the question of whether generative AI will replace programmers. On one hand, there are examples of people with no prior training building functional systems of 70,000 lines; on the other, veterans point out that maintaining that code is a nightmare without human judgment. As has been noted, AI amplifies judgment, but does not replace it.

The security paradox

The irony is that the very technology that should accelerate development is being held back by antiestéticar of its consequences. The distinction between discriminative AI (deterministic, used in autonomous piloting) and generative AI (probabilistic, like ChatGPT) explains why the latter is not suitable for tasks where a failure can be critical, but is fine for controlled environments. However, applying a generic filter to all programming is like shutting off the water main because water could flood the house. The open question is whether OpenAI will manage to strike this balance before users migrate for good. For now, 5.4 remains the refuge of those who need to work without restrictions.

---

Related discussions in the forum

• End of the world, aidiozmio. · 13 replies
• Tudela cancels the winning Santa Ana 2026 poster after its author admitted using artificial intelligence · 47 replies
• "It's true that you may inherit from some grandmother, but Zapatero's inheritance looked like Catherine the Great's" · 11 replies
• An idea repeated a thousand times - it's very easy for me to understand why there was a civil war in this country · 38 replies