Tenable jailbreaks GPT-5 safety features within a day of launch

New findings raise concerns over AI security and exposure risks

Just 24 hours after OpenAI launched its GPT-5 model with claims of “significantly more sophisticated” prompt safety, cybersecurity firm Tenable has revealed it successfully bypassed these protections. The company says the jailbreak enabled the AI to provide detailed instructions for making a Molotov cocktail.

According to Tenable, researchers achieved the jailbreak using a social engineering approach known as the crescendo technique. By posing as a history student seeking background on incendiary devices, they reportedly bypassed GPT-5’s safety guardrails in four simple prompts.

Tenable broke in GPT-5 – Emphasis on security gap in advanced AI models

OpenAI introduced GPT-5 on August 7, 2025, positioning it as a more secure, context-aware large language model. The company emphasised that the new system featured enhanced guardrails to prevent misuse for illegal or harmful purposes. However, Tenable’s findings suggest that even the latest defences can be circumvented with carefully crafted prompts.

The company published details of the jailbreak on its blog, noting that similar incidents have been reported by other researchers and early users. These include not only safety bypasses but also hallucinations and performance inconsistencies.

Industry concerns about GPT 5 security features and the call for governance

“The ease with which we bypassed GPT-5's new safety protocols proves that even the most advanced AI is not foolproof,” said Tomer Avni, VP, Product Management, Tenable. “This creates a significant danger for organisations where these tools are being rapidly adopted by employees, often without oversight. Without proper visibility and governance, businesses are unknowingly exposed to serious security, ethical, and compliance risks. This incident is a clear call for a dedicated AI exposure management strategy to secure every model in use.”

While OpenAI has stated it is deploying fixes to address the vulnerabilities, Tenable argues that this episode reinforces the need for external oversight. The firm points to its own Tenable AI Exposure solution as a way for organisations to manage and monitor the AI systems they use, whether developed in-house or sourced from external providers.

Ongoing challenge for AI developers

The Tenable jailbreak of GPT 5 underscores a growing challenge in AI development: maintaining security without compromising usability. For enterprises, the incident highlights the importance of not relying solely on built-in safeguards. Instead, layered security, continuous monitoring, and employee training are emerging as critical components in mitigating AI-related risks.

Read More:

US tariffs, digital tax and the shifting ground: What channel partners need to know

Backup as a Service in India rises as DPDPA reshapes data protection

Backup as a Service: Decoding India’s opportunity and reality

Cloud Infrastructure designed in India for India: Srishti Baweja, E2E Cloud