PraisonAI before version 4.6.78 has an authentication bypass that allows a remote attacker to list and invoke agents without authentication. The vulnerability arises when the PRAISONAI_CALL_AUTH setting is disabled; the application mistakenly trusts the client-supplied Host header to determine whether the connection is local. By sending a spoofed Host header such as "127.0.0.1", a non-privileged attacker can trick the service into treating the request as local and therefore bypass the authentication check. This weakness falls under the authentication bypass category (CWE-287). The consequence is the ability to perform all agent operations exposed through the /api/v1/agents endpoints without credentials.

The affected product is MervinPraison PraisonAI. All releases prior to version 4.6.78 are vulnerable; versions 4.6.78 and later contain the fix as indicated by the commit information listed in the references.

The CVSS score of 8.8 indicates a high severity risk. The EPSS score is reported as < 1%, meaning exploitation attempts are currently very infrequent, and the vulnerability is not listed in the CISA KEV catalog. A remote attacker who can reach the service over the network can exploit the flaw by sending a crafted Host header. No special privileges or additional access are required beyond network connectivity to the target system.