Typebot CVE-2026-39967

Four simultaneous conditions must be met. First, the attacker must be authenticated to the Typebot instance (CVSS PR:L). Second, the target typebot must have the rememberUser feature enabled - without it, resultId-based session resumption does not occur and the vulnerable code path is never reached. Third, the attacker must possess a valid, non-guessable CUID2 resultId from a result record owned by a different typebot; brute-force is explicitly infeasible due to the 24-character cryptographically random format, so the ID must be obtained through another means. Fourth, the attacker's typebot must define variable names that overlap with those in the victim's typebot for any data to surface. What limits exploitation: the CUID2 ID space makes speculative attacks impractical; the rememberUser requirement means default-configured typebots may not be reachable; and authentication bars anonymous exploitation.

The CVSS 3.1 score of 3.1 (Low) is consistent with the signal profile here: AV:N (network reachable) is offset by AC:H (high complexity), PR:L (requires authenticated attacker), and C:L (limited confidentiality impact with no integrity or availability consequences). The AC:H rating is directly justified by the CUID2 constraint - the 24-character cryptographically random ID makes brute-force infeasible, meaning the attacker must independently obtain a valid resultId through some out-of-band means (e.g., insider access, log exposure, or prior compromise of another system component). No EPSS score was provided, and the vulnerability is not listed in CISA KEV, aligning with the low-severity rating. There is no evidence of public exploit code. In practice, real-world risk is narrow: the attacker must be authenticated, must have or guess a valid CUID2 resultId belonging to another typebot, and the target typebot must have rememberUser enabled with variable names overlapping those of the attacker's typebot. For multi-tenant SaaS deployments of Typebot where untrusted users can create bots on the same instance, the risk is elevated despite the low CVSS - a malicious tenant could potentially access PII from another tenant's users if they obtain a resultId.

An authenticated user on a shared Typebot instance creates their own typebot and somehow obtains a valid CUID2 resultId belonging to a different user's typebot (e.g., through an exposed server log, a support ticket, or another compromised component). They call the startChat endpoint on their own typebot, supplying the foreign resultId. Because the pre-patch findResult query does not check typebotId, the database returns the victim's result record, and if variable names overlap, the attacker's session is populated with the victim's prior answers and PII. No public proof-of-concept code exists at time of analysis.