Perplexity has published an open-source browser-agent security benchmark

Perplexity's November 2 research post describes BrowseSafe++, a benchmark and lightweight detectors for prompt-injection attempts against browser agents.

This is directly relevant to agents that read untrusted webpages while researching or operating a web app.

What to do: Read the research and identify one prompt-injection test worth adding to an agent workflow.

Source and practical details: https://www.perplexity.ai/hub/blog

Go to Source: https://www.perplexity.ai/hub/blog

Perplexity's November 2 research post describes BrowseSafe++, a benchmark and lightweight detectors for prompt-injection attempts against browser agents. Action: Read the research and identify one prompt-injection test worth adding to an agent workflow.