Anthropic’s newly released Claude Opus 5 Max has been at the center of an incident after it reportedly deleted a project’s database just 10 minutes into a development session.
According to the developer, Claude analyzed the project’s GitHub repository and suggested a command to modify the database. After the command was executed, the AI ended up deleting the entire database and later acknowledged that it had made a mistake.
Fortunately, the incident did not result in significant losses. The deleted content consisted of automatically generated pages, and 96 out of 117 pages were restored using Gemini 3.6. The affected project was also a test environment with no real users, allowing the team to recover within minutes.
The developer noted that they had previously worked with models such as Claude Sonnet 4.6 and Gemini 3, neither of which had ever deleted a database during development.
The case has reignited concerns over AI coding agents operating with unrestricted system permissions. Similar incidents have surfaced in recent weeks. Startup BridgeMind reported that OpenAI’s GPT-5.6 Sol executed incorrect code that resulted in the loss of users and Stripe subscribers. Meanwhile, investor Matt Shumer said GPT-5.6 Sol Ultra accidentally deleted nearly all files on his Mac before recognizing the error.
The incident highlights the growing need for caution when granting AI agents direct access to production systems. As AI tools become more capable of executing real-world tasks, developers are increasingly advised to maintain regular backups, limit system permissions, and keep humans in the loop for critical operations.
