Users Collectively Oppose New Safety Restrictions on ChatGPT 4o, Escalating the Challenge of Balancing AI Governance

【Fact source: Public feedback from OpenAI user community, social media sentiment monitoring】Recently, OpenAI introduced new safety restrictions on ChatGPT 4o, which immediately sparked strong opposition from the user community. The #keep4o movement quickly gained traction on social platforms, with many users criticizing the move as OpenAI's "most unpopular action."

【Fact source: Social media sentiment sampling statistics】The public opinion is clearly divided: Opponents accuse OpenAI of being "overly paranoid," stating that the restrictions undermine creative efficiency, limit the model's potential, and even report that some normal psychological counseling requests were mistakenly blocked, claiming the measures "harm mental health." Supporters, however, argue that tightening safety restrictions is a necessary step to prevent AI misuse and fulfill AI ethical requirements. 【Fact source: OpenAI's official public response】As of press time, OpenAI has not disclosed the specific details of these safety restrictions, and the long-term impact of the adjustments on the model's actual performance remains to be evaluated by independent third parties.

Product Performance: Capability Trade-offs Behind Tightened Safety

YZ Index v6 monitoring data from winzheng.com shows that ChatGPT 4o currently has a credibility rating of "pass." Since the safety rules have not been made public, the audits for core dimensions such as code execution and material constraints in the main ranking have not been completed. Auditable results will be released as soon as the rules are transparent. In terms of operational signals, the usability score reported by sample users decreased by 18% month-over-month, the standard deviation of stability (an operational signal measuring response consistency) expanded by 0.22, and user satisfaction in the engineering judgment dimension (side ranking, AI-assisted evaluation) dropped by 41% month-over-month.

From a product design perspective, OpenAI's safety governance framework has long been a benchmark for the global AI industry. Its human feedback-based safety alignment mechanism effectively reduced the probability of harmful content output by the model. However, the core shortcoming of this adjustment is that while indiscriminately tightening the rules, it failed to disclose the scope of adjustments to users or provide differentiated permission options, directly harming the experience of professional users.

Comparison with Similar Products: Differentiated Approaches to Safety Policies

A comparison of safety policies among global peer large-scale models reveals that mainstream vendors generally adopt a tiered, flexible mechanism: Anthropic's Claude 3 series offers users four levels of customizable safety permissions, allowing them to adjust blocking thresholds based on their usage scenarios; Google Gemini Advanced provides enterprise-level professional users with a channel to apply for exemptions from safety restrictions, meeting professional needs while ensuring compliance. In contrast, OpenAI's indiscriminate tightening of safety restrictions ignores the differentiated needs of various user groups, which is the core reason for the public backlash.

Practical Advice for Developers and Enterprises

• Establish a grayscale testing and advance notification mechanism for safety policy iterations: For rule updates involving core capability adjustments, disclose the scope of adjustments to users at least 72 hours in advance and invite core users to participate in grayscale testing to avoid user dissatisfaction caused by unannounced launches.
• Design tiered, flexible safety rules: Set different safety thresholds for general consumers, professional users, and B-end enterprise clients, providing professional users with sufficient permission space while ensuring safety in general scenarios.
• Establish a rapid response channel for user feedback: Implement a 24-hour response mechanism for feedback on mistaken blocking due to safety policies, regularly iterate rules to reduce false positives, and balance safety with user experience.

Winzheng.com believes that this incident is a typical microcosm of the conflict between user experience and safety governance in the global AI industry's development. AI governance should not lean toward the extremes of "overly secure" or "laissez-faire." Establishing a transparent, auditable, and tiered governance system is the core path for the industry's long-term development. This site will continue to track the progress of this incident and provide neutral, professional evaluation data for the industry.