Replit adds black-box pen tests that attack apps like an outside hacker

Archive entry: this covers an event from August 17, 2026, and we published it on November 8, 2026. The details come from sources dated at the time and may have changed since.

On August 17, 2026, Replit added black-box penetration tests, which test an app the way an external attacker would: over the network, through a browser, with no access to source code. The agent-driven scans ran against a sandboxed copy of the app, alongside Replit's existing white-box code scans, as described in Black-box pen tests on Replit.

How the scans worked

• Level 3 scan: started from the project's Security Center, it ran a white-box scan and a black-box scan in parallel, and the black-box scanner got only the app's link.
• Sandbox: each scan ran against a full copy of the app in a private sandbox, so nothing it tried could escape or reach the app's users.
• Test sequence: the agent first tested as an anonymous visitor, then as an ordinary signed-in user, checking whether it could open other users' records or reach admin-only areas.
• Results: the agent returned a list of confirmed problems to fix with Replit Agent.

What it means for people building with AI

Pre-launch penetration testing became a built-in step for apps built with AI on Replit. Replit said findings from the two scan types rarely overlapped, and in one case only the black-box scan found an admin dashboard at a guessable address with no login.

Scan levels

Level 1 ran dependency checks and static analysis and was free. Level 2 ran the white-box agent scan, and Level 3 ran both scanners.

Sources

• Black-box pen tests on Replit (Replit, August 17, 2026, official)

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.