AI News: ChatGPT-5 Vulnerability Lets Hackers Bypass AI Security With Simple Prompts

CopilotPerplexityChatGPTDeepSeekGoogleGrokGemini

A newly reported flaw in OpenAI’s flagship model, ChatGPT-5, has raised concerns about the safety of advanced artificial intelligence systems. According to cybersecurity experts, the vulnerability allows attackers to bypass security measures by using carefully crafted phrases.

The ‘PROMISQROUTE’ Weakness

The issue, identified as “PROMISQROUTE,” is linked to the way AI services manage computational resources. Instead of always using the most advanced model for every request, background systems known as “routers” decide which model will handle a user’s input. This approach helps reduce costs and balance performance, but researchers say it opens a pathway for attackers to exploit weaker models within the system.

How the Exploit Works

When a user enters a prompt, the router may direct it to one of several models in a shared “model zoo.” Experts explain that attackers can manipulate prompts so that the request avoids ChatGPT-5’s stricter safeguards and gets processed by a less-protected model. Once this happens, the AI may generate responses that bypass intended restrictions.

Impact on AI Safety

The discovery highlights a broader challenge facing AI providers: maintaining strong security across multiple interconnected models. While advanced versions like ChatGPT-5 are equipped with stronger guardrails, the reliance on a mixed-model infrastructure can create gaps. Security researchers warn that attackers may take advantage of these inconsistencies to produce harmful outputs or extract sensitive information.

The report has sparked calls for AI vendors to tighten their routing systems and ensure that safety standards remain consistent across all deployed models. Analysts say the issue is not unique to OpenAI, as other major AI providers also use similar cost-saving architectures to handle the heavy demand for large-scale AI services.

The disclosure of the PROMISQROUTE flaw underlines the growing importance of cybersecurity in the AI sector, as advanced models become central to business, education, and public use worldwide.