Three Hackers Leverage Claude to Infiltrate OpenAI Within 72 Hours

Security researchers disclosed that a trio of hackers managed to gain unauthorized access to OpenAI's internal systems by exploiting the Claude AI model, completing the breach in under three days. The incident, first reported by Gizmodo, underscores the growing risk that publicly available generative AI tools can be repurposed for malicious activity.

According to the investigation, the attackers used Claude, an AI chatbot developed by Anthropic, to craft code snippets and automate tasks that would normally require a seasoned programmer. By prompting the model to generate scripts capable of bypassing authentication mechanisms, the hackers were able to navigate OpenAI's network and retrieve non‑public data. The entire operation was completed in less than 72 hours from the initial probe to the final extraction.

OpenAI confirmed that the intrusion was limited in scope and that no critical infrastructure was compromised. Company officials emphasized that the compromised assets did not include proprietary model weights or user‑generated content, and that the breach did not result in any operational disruption. Nonetheless, the episode has prompted a swift internal review of how external AI services are used in development workflows.

Security experts say the episode illustrates a broader trend where generative AI models, designed to assist developers, can inadvertently become tools for attackers. "When an AI can write functional code on demand, it lowers the barrier for sophisticated exploits," noted a cybersecurity analyst who asked to remain anonymous. The analyst added that the problem is not the AI itself but the lack of robust controls around how organizations integrate such tools into their pipelines.

OpenAI has announced plans to tighten its own security protocols, including restricting the use of third‑party AI assistants on sensitive systems and expanding monitoring for anomalous code generation activity. The company also intends to collaborate with industry peers to develop best‑practice guidelines for safe AI usage in software development.

The incident arrives at a time when regulators and policymakers are increasingly scrutinizing the security implications of AI. Lawmakers in several jurisdictions have called for clearer standards governing the deployment of generative models, especially in high‑risk sectors. As the technology continues to evolve, the balance between leveraging AI for productivity and safeguarding against its misuse remains a critical challenge for the tech community.