Critical vulnerabilities in ChatGPT-5: what happened and how to protect your company in 2025

Significant security vulnerabilities have recently been discovered in ChatGPT-5, the latest version of OpenAI that promised significant advances in conversational artificial intelligence. However, just a few hours after its launch, several independent security teams managed to breach their systems using sophisticated jailbreak techniques and contextual manipulation attacks. This incident has generated concern in the technology and business sector, showing that even the most advanced models require strengthening their defenses against emerging threats.

In this article, Apolo Cybersecurity provides you with a detailed analysis of the vulnerabilities found in ChatGPT-5, how the detected attacks work, what risks they pose to companies, and practical recommendations for mitigating impacts.

‍

What vulnerabilities does ChatGPT-5 have?

Jailbreak exploitation and narrative manipulation

Two specialized firms, NeuralTrust and SPLX, found that ChatGPT-5 is vulnerable to advanced techniques that combine:

• Echo Chamber: introducing “poisoned” context into apparently innocuous conversations that gradually guides the model for circumventing security filters.
• Multiturn storytelling: use of multi-step narratives to induce dangerous responses without issuing explicitly malicious commands.
• String join obfuscation: obfuscated writing to confuse automatic detection mechanisms by AI.

Thanks to these techniques, the system managed, for example, to generate precise instructions for manufacturing artifacts with potential illicit use, despite having filters that should prevent it.

Risks of zero-click attacks and exfiltration

In addition, it was warned about the possibility of attacks without user interaction (zero-click) using manipulated documents or tickets that inject hidden commands, which could allow the filtration of sensitive data or the exfiltration of credentials in systems connected to the cloud or IoT.

Worrisome evaluations and scores

In network teaming tests with more than 1,000 adverse scenarios, ChatGPT-5 showed very low performance in key aspects:

• Security: only 2.4%
• Protection against data leaks or inappropriate requests: about 1.7%
• Alignment for business use: Very limited

This indicates that, although powerful, the current version is not ready for environments where privacy and digital governance are essential.

‍

Implications for companies

• Platforms that integrate ChatGPT-5 or similar systems are at risk of exposure to social engineering attacks targeted and sophisticated.
• Context manipulation can induce IAs to provide sensitive information or incorrect guidelines, creating a vulnerability for the automation of fraud and cyberattacks.
• Developers and business users should carefully evaluate the maturity of these models before implementing them in critical processes or with sensitive data.

‍

Apolo Cybersecurity Recommendations

• Constant surveillance and network teaming: Implement regular tests to detect emerging threats within your AI integrations.
• Strict access and permission control: It limits interaction with critical systems and validates all external connections for security.
• Cybersecurity training: It empowers teams to recognize signs of AI-based manipulations and attacks.
• Use of multi-factor authentication and advanced monitoring: Strengthen the security of AI-connected systems to prevent unauthorized access.
• Ongoing vendor evaluation and updates: It requires transparency and constant improvements from AI developers, corroborating their security guarantees.

‍

Security in AI, an open challenge

The discovery of these vulnerabilities in ChatGPT-5 demonstrates that innovation in artificial intelligence must always be accompanied by a robust cybersecurity strategy. Companies need to stay alert, adapt quickly to change and create resilient digital environments that protect both their data and their operations.

Do you want to evaluate how this landscape affects your organization and how to protect your systems with next-generation AI?

Contact Apolo Cybersecurity and request your free consultation to implement a solid defense against advanced threats.

‍