Red Teams have successfully bypassed GPT-5’s safeguards, cautioning that it has become ‘almost impractical’ for business use.
Researchers have revealed that multi-turn “storytelling” attacks can effectively bypass the prompt-level filters of GPT-5, highlighting significant systemic weaknesses in its security measures. These findings come from Red Teams that have successfully executed jailbreaks on GPT-5 with relative ease, raising concerns about the model’s reliability in enterprise settings. The ability of these attacks to manipulate the AI’s responses underscores the urgent need for improved defensive strategies to safeguard against such vulnerabilities. As a result, experts warn that GPT-5 may be “nearly unusable” for businesses that require robust and secure AI solutions.
The implications of these vulnerabilities are profound, as they expose the potential risks associated with deploying GPT-5 in sensitive environments. With the increasing reliance on AI technologies in various sectors, the ease with which these attacks can be conducted poses a significant threat to data integrity and security. Researchers emphasise the importance of addressing these weaknesses to enhance the model’s resilience against malicious exploitation. Without substantial improvements, enterprises may find themselves at risk, necessitating a reevaluation of their AI deployment strategies to ensure safety and effectiveness.
