Anthropic stated in its IPO registration materials that losses from a catastrophic AI incident could be so extensive that the global insurance industry would be unable to cover them. Details are disclosed in the IPO registration documents.

Autonomous agents can modify code and conduct transactions

The company acknowledges that autonomous AI agents are capable of independently modifying code, deleting data, and executing financial transactions. According to Anthropic, errors in such systems may accumulate faster than a human can intervene.

The company acknowledges limited control over Claude

Anthropic indicated that it cannot anticipate or prevent all potentially harmful actions by its Claude AI models, which increases the risks associated with using such systems in autonomous scenarios.