CodexBar CVE-2026-49949

Exploitation requires the attacker to be network-adjacent and capable of injecting or substituting HTTP redirect responses in the path between CodexBar and the targeted AI provider backend - achievable via ARP poisoning, a rogue access point, or a compromised upstream proxy on the local network segment. The victim must actively initiate a credentialed provider request (UI:P per CVSS 4.0), meaning exploitation cannot occur passively without user action. Attack complexity is HIGH (AC:H), ruling out opportunistic internet-scale exploitation. No attacker authentication to CodexBar or to the provider is required. Exploitation is further limited to environments where network-adjacent positioning is feasible: shared Wi-Fi networks, local enterprise segments without proper network isolation, or scenarios involving a compromised intermediary proxy. Same-origin HTTPS redirects are NOT vulnerable - only redirects that cross to a different origin or downgrade to plaintext HTTP trigger credential forwarding.

The CVSS 4.0 score of 6.0 (AV:N/AC:H/AT:N/PR:N/UI:P/VC:H/VI:N/VA:N/SC:N/SI:N/SA:N) accurately reflects a meaningful but exploitation-constrained credential theft risk. Attack complexity HIGH (AC:H) is the most significant limiting factor - the attacker must occupy a network-adjacent MITM position capable of injecting redirect responses between CodexBar and the AI provider backend, which eliminates purely opportunistic remote exploitation. User interaction (UI:P) is required, meaning the victim must actively trigger a provider request. Confidentiality impact is HIGH (VC:H) because successfully captured API keys or bearer tokens can grant persistent unauthorized access to the victim's AI provider accounts - an impact disproportionate to the raw score in high-value environments. Integrity and availability are unaffected. No CISA KEV listing exists, and no public proof-of-concept code has been identified, indicating no observed widespread exploitation. EPSS data was not provided in the available intelligence. The real-world priority is moderate-to-high for users operating CodexBar on shared, corporate, or untrusted Wi-Fi networks where MITM positioning is more achievable, and lower for users on isolated home networks. The absence of a scope change (SC/SI/SA all N) confirms impact is limited to the vulnerable system's own credentials.

An attacker on the same network segment as a CodexBar user - for example, on shared office or café Wi-Fi - establishes a MITM position via ARP spoofing or a rogue access point. When the user triggers a Copilot or Vertex AI request through CodexBar, the attacker intercepts the AI provider's HTTP response and replaces it with a 301/302 redirect pointing to a plaintext HTTP endpoint under the attacker's control. CodexBar's unpatched ProviderHTTPClient follows the redirect and reattaches the original Authorization header or session cookies to the new request, delivering the credentials directly to the attacker. No public exploit code has been identified at time of analysis.