OpenClaw session tools – sessions_list, sessions_history, sessions_send – were able to target and access higher‑level sessions than intended in shared‑agent configurations, exposing transcript content across peer sessions. Additionally, in Telegram webhook mode the system did not fall back to a per‑account `webhookSecret` when only an account‑level secret existed, which could allow an attacker to intercept Telegram messages if the wrong secret is used. The underlying weaknesses correspond to CWE‑209 (information exposure) and CWE‑346 (broken access control).

The vulnerability affects OpenClaw deployments running any version older than 2026.2.15, especially those configured for shared‑agent, multi‑user scenarios. It is also relevant for deployments using Telegram webhook monitoring without an explicit per‑account webhook secret override. Vendors and product owners should verify that their instances fall into this older version range.

The CVSS v3.1 base score is 6.9, reflecting moderate severity. EPSS indicates exploitation probability is below 1 % and the flaw is not listed in the CISA KEV catalog, suggesting it is not currently known to be widely exploited. Inferred attack vector is a local or network attacker who can join or interact in a shared‑agent session or who can manipulate the Telegram webhook configuration in a trusted account. Exploitation requires the attacker to be present in the same shared‑agent context or to control the Telegram monitoring endpoint, after which the attacker may read or interfere with session transcripts or compromise message handling.