OpenAI Threatening to Ban Users for Asking Strawberry About Its Reasoning

Hello!

OpenAI claims that its latest AI model, code-named "Strawberry" and released as o1-preview, is capable of "reasoning." Yet understanding how its thought process works remains something the ChatGPT maker is determined to keep off-limits.

As Ars Technica reports, OpenAI is now threatening to ban users who attempt to coax the large language model into revealing its reasoning — a clear sign of how far the company has moved from its original open-source AI ethos.

Policy Warnings

According to social media accounts, users are receiving emails from the Microsoft-backed startup stating that their requests to ChatGPT have been flagged for "attempting to circumvent safeguards."

"Additional violations of this policy may result in loss of access to GPT-4o with Reasoning," the emails warn.

Hush Hush

The clampdown is especially ironic given that much of the hype around Strawberry centered on its "chain-of-thought" reasoning, which lets the AI articulate its step-by-step path to an answer. OpenAI chief technology officer Mira Murati described this capability as a "new paradigm" for the technology.

Reports differ on exactly what triggers the violations. As Ars discovered, some users say the phrase "reasoning trace" is enough to cause trouble, while others report that simply using the word "reasoning" activates OpenAI’s automated systems.

Users can still view a summary of Strawberry’s thought process, but it is generated by a second AI model and significantly condensed.

In a blog post, OpenAI explains that hiding the full chain-of-thought allows it to avoid applying safety filters to the model’s internal reasoning. This approach lets developers safely examine the AI’s “raw” thought process without exposing potentially non-compliant content. The company also acknowledges that the restriction helps preserve its "competitive advantage" by preventing rivals from replicating its methods.

Red Alert

The downside is that this policy concentrates responsibility for model alignment firmly in OpenAI’s hands rather than encouraging broader transparency. The change creates challenges for red-teamers — researchers who probe AI systems to identify and fix safety issues.

"I’m not at all happy about this policy decision," AI researcher Simon Willison wrote on his blog, as quoted by Ars.

"As someone who develops against LLMs, interpretability and transparency are everything to me — the idea that I can run a complex prompt and have key details of how that prompt was evaluated hidden from me feels like a big step backwards."

OpenAI appears to be moving further toward keeping its AI models as opaque black boxes.

Also read:

• Online Courses for Elevating Your Business Skills
• Why proxy providers can be a safer option than VPNs
• 10 Best Open-source Business intelligence Tools

Thank you!
Join us on social media!
See you!