Hacktron Uses Anthropic's Claude to Breach OpenAI Systems in 72 Hours

Hacktron successfully exploited Anthropic's Claude, specifically version Opus 4.8, to breach OpenAI's internal systems within 72 hours. The total cost for the AI resources used in the project was $3,000. The attack was adapted to new targets in just 1 to 2 days after the initial breach. OpenAI took 14 hours to confirm a fix after the exploit was executed, which took Claude only 4 hours to gain control of the server. This incident highlights a significant shift in cybersecurity capabilities, as Hacktron noted that tasks requiring extensive resources and time can now be completed in days. This follows the recent release of Claude Opus 5 on July 24, 2026, which may have implications for the security of AI systems reliant on similar models.