Opus 5 may have solved browser-based prompt injection, the biggest security flaw haunting AI agents

Anthropic says Opus 5 nearly eliminates browser-based prompt injection in its own products when Auto Mode is enabled.

The company reported zero successful attacks in 129 test cases, while independent tests also found lower injection rates than earlier models.

The improvement could strengthen the safety of AI browser agents, which are especially vulnerable to hidden instructions on web pages.

But the model is not fully immune without additional protections, so prompt injection remains a live security risk.