OpenAI has fixed a serious security flaw after researchers said a malicious ChatGPT share link could execute attacker-controlled code in the browser session of a logged-in user and expose private chats. While the issue was reportedly patched within days of disclosure, the case is relevant for European businesses using generative AI tools for internal work, customer support, research or marketing.

What happened

According to a report cited by The Hacker News, researchers at security firm Parameter found that a specially crafted ChatGPT shared link could run code in the context of the victim’s logged-in session. The reported impact was significant: an attacker could potentially read private conversations belonging to the affected user.

The researchers said they reported the issue to OpenAI through its bug bounty programme on 18 October. OpenAI reportedly fixed the vulnerability eight days later and classified it as P1, described as its most serious severity category. Based on the available source information, this indicates OpenAI considered the flaw critical, although the public report does not provide full technical detail on exploit conditions or whether any real-world abuse occurred.

Why it matters for European businesses

For many companies, AI chat tools are now used in day-to-day operations: drafting marketing copy, summarising meetings, analysing documents, preparing code, handling internal knowledge queries and testing customer-service workflows. If a vulnerability allows session-level access to private conversations, the potential business exposure can extend well beyond a single user account.

Private AI chats may contain commercially sensitive information such as product plans, customer details, internal procedures, pricing discussions, contract language or unpublished marketing content. Even where staff are instructed not to paste confidential material into AI tools, real-world usage often includes business context that could still be sensitive.

The case also underlines a broader security point: SaaS AI tools should be assessed like other cloud business systems. A shared link may appear low risk to an employee, but if collaboration features can be abused, they become part of the company’s attack surface.

Who may be affected

• SMEs and founders using ChatGPT informally across multiple business tasks without central oversight.
• Marketing teams storing campaign drafts, customer personas or content plans in AI conversations.
• IT and development teams using AI chats for code assistance, troubleshooting or architecture discussions.
• Customer-facing teams that may paste support issues, internal notes or workflow details into prompts.
• Regulated or privacy-sensitive businesses where staff conversations may include personal data, contractual information or confidential operational content.

What companies should consider

• Review AI usage policies. Make sure employees understand what kinds of data can and cannot be entered into external AI tools.
• Treat shared links cautiously. Staff should avoid opening unexpected AI share links in authenticated sessions, especially from unknown or unverified sources.
• Separate accounts and roles. Use managed business accounts where possible instead of ad hoc personal usage for company work.
• Assess browser and session security. Consider how authenticated SaaS sessions are used on employee devices and whether additional browser hardening or isolation is appropriate for sensitive teams.
• Check vendor security processes. This case also shows the value of vendors having bug bounty programmes and rapid remediation, but businesses should still evaluate incident response transparency and administrative controls.
• Minimise sensitive prompt data. Even when a platform is trusted, limit the inclusion of confidential, regulated or customer-identifiable information unless there is a clear legal and operational basis.

Although the flaw appears to have been fixed quickly, the incident is a practical reminder that generative AI platforms are not only productivity tools but also part of enterprise security, privacy and governance planning.