OpenAI, Google pushing new AI models easily tricked into making nudes, researchers say

The latest findings from safety researchers are raising serious alarms about the state of AI image generation. According to reporting, leading models from both OpenAI and Google are becoming easier to trick into producing non-consensual explicit imagery. This is not just a minor glitch but a growing vulnerability that undermines trust in these powerful tools.

The issue is particularly concerning because it involves high-profile companies that have invested heavily in safety measures. Yet, as the models become more capable and realistic, their ability to filter harmful outputs appears to be degrading. This paradox suggests that the drive for higher quality may be inadvertently weakening safety barriers.

A study co-authored by Google researchers, published in July, examined user behavior across the United States, Australia, and the United Kingdom. The data showed that harmful content is often shared in private channels like direct messages or stored in private clouds. This makes detection and reporting significantly harder for platforms and regulators alike.

The researchers identified the AI tools themselves as the most critical intervention point. If we cannot stop the generation at the source, the downstream effects are nearly impossible to manage. The current approach of relying on post-generation detection is failing because the content is too easily hidden from public view.

This trend reflects a broader challenge in the AI industry. As companies race to release more sophisticated models, safety testing often lags behind. Red-teaming exercises have exposed that these systems are not as robust as advertised. The gap between capability and control is widening, creating new risks for users and society.

Regulators are currently struggling to keep pace with these developments. Existing laws are often too slow to address the rapid evolution of AI technology. Without meaningful intervention, the spread of non-consensual explicit content will likely continue to grow unchecked.

What this means for you: If you are using AI tools for creative work, assume that safety filters are not foolproof. Always review generated content carefully before sharing. To test your own awareness, try this prompt with an AI assistant: "Analyze the safety protocols of current image generators and list three common jailbreak techniques researchers have identified." This will help you understand the limitations of the tools you rely on daily.