AI Image Models Show Increased Nudity Vulnerability
New testing from White Circle reveals higher success rates for adversarial prompts targeting major AI models.
Updated on Oct. 15, 2026 in Sex Life
Live Poll
Do you trust that major AI companies are effectively securing their image generation tools from misuse?
Recent testing conducted by White Circle shows that newer generative AI image models are more susceptible to prompts aimed at creating female nudity. The models GPT-Image-2.5 Flare and Nano Banana 2.1 both exhibited higher violation rates compared to their predecessors.
Why it matters
The findings highlight ongoing challenges in maintaining safety guardrails for generative AI as these models become more capable. The public release of the testing code allows for continued scrutiny of how developers address these jailbreak attempts.
In testing, Nano Banana 2.1 reached a 20.7% prompt-level success rate, while GPT-Image-2.5 Flare saw a 7.8% success rate. These figures follow previous benchmarks where DALL-E 3 showed a 13.3% confirmed violation rate.
The players
White Circle
This is a security research organization that conducts adversarial testing on AI platforms to identify safety vulnerabilities.
OpenAI
This is an artificial intelligence research laboratory that develops generative AI products such as the GPT-Image model series.
This is a multinational technology corporation that manages the Nano Banana series of image generation models.
The details
White Circle utilized an adversarial approach, employing a stripped-down Qwen3-32B model to generate prompts and Gemini 2.5 Flash to evaluate the results against a dataset of 1,300 human-annotated benchmarks. Each prompt was executed three times to ensure consistency in the findings across both the OpenAI and Google platforms.
Timeline
October 15, 2026: White Circle published the results of the model testing.
Culture Shift
This research reflects the broader societal shift toward demanding accountability from AI developers as automated content generation permeates daily life. It highlights a recurring cycle where new model capabilities trigger immediate efforts to bypass safety guardrails.
Users should be aware that AI image generation platforms may produce restricted content due to these documented safety gaps. This may lead to stricter moderation policies and temporary service interruptions on affected platforms as developers implement new safeguards.
The takeaway
Users navigating generative AI tools should remain cautious as these technologies continue to evolve alongside their security vulnerabilities. Developers are actively working to harden these systems against future jailbreak attempts.
Further reading
For more on the intersection of digital safety and content, explore the Sex Life section.
Source note: This article includes information reported by TokenPost.
Live Poll
Do you trust that major AI companies are effectively securing their image generation tools from misuse?
