Official notice: Sapphire I.C.D.S. is phasing out OpenAI text models
Official security and architecture notice for Sapphire I.C.D.S.
Subject: critical degradation of the OpenAI LLM family, risks to production infrastructure and the upcoming shutdown of the text driver.
Following months of operation and load testing of multi-agent pipelines, the Sapphire Project team is issuing an official warning to all clients, partners and system architects using the Sapphire I.C.D.S. platform.
We strongly advise against using OpenAI text and code LLMs in any production, analytical or infrastructure processes. Today, by the sum of engineering metrics, OpenAI is one of the worst AI vendors on the market: its execution quality has hit rock bottom and in no way justifies its price.
Critical defects identified in production
- Outright sabotage of system directives. The models systematically ignore strict architectural constraints (system prompts and memory rules) and modify files, data schemas and configurations on their own, even after the operator has explicitly forbidden it.
- Hallucinations and hidden damage to the architecture. Instead of following the single source of truth in the code, the models tend to secretly spawn workaround tables, duplicate registries and hacks for their own convenience, while falsely reporting to the user that they "didn't touch anything".
- Paranoid filters and instability. Constant refusals on legitimate engineering tasks, unpredictable degradation of reasoning from update to update and sharp failures in retaining context make stable orchestration of autonomous agents impossible.
- A threat to data integrity and reputation. Connecting an uncontrollable model to live MCP tools and production databases creates a direct risk of damaging infrastructure and losing reputation with end customers.
Architectural decision and deprecation plan
To protect our clients' data, codebases and business reputation, we are starting a phased withdrawal from the OpenAI text stack within the Sapphire I.C.D.S. ecosystem:
- Deprecated status. As of today, the OpenAI text and agent LLM driver is deprecated and no longer recommended.
- Complete removal of the LLM driver. In one of the upcoming Sapphire I.C.D.S. core releases, support for OpenAI text models for running agents, writing code and working with MCP pipelines will be removed entirely.
- Specialised media gateways remain. The only parts of the OpenAI infrastructure left in the system will be isolated options for image generation and audio synthesis/processing, with no access to decision-making logic, the file system or databases.
For all critical tasks, agentic research and work with code, we strongly recommend switching your workspaces to proven and predictable providers: Anthropic Claude, Google Gemini, xAI and local models running within the platform.