Attack ChainAIDerived
Hypothetical attack flow derived from CVE metadata – not a confirmed exploit chain
-
AccessAuthenticate as low-privilege tenant user
-
DeliverySubmit code via POST /v1/tools/run or custom tool
-
ExploitLocalSandbox subprocess inherits full server env
-
ExecutionRead LETTA_PG_URI, POSTGRES_PASSWORD and API keys
-
ImpactConnect to shared Postgres and read cross-tenant data
Vulnerability Assessment
| Weakness |
CWE
OWASP Top 10:2025
A02:2025 - Security Misconfiguration
ASVS 5.0
V13 - Configuration Environment variable protection
|
| Exploitation Conditions | Requires (1) an authenticated account with privileges to execute tools - the attacker must be able to call POST /v1/tools/run or register/run a custom tool (PR:L, so not anonymous); and (2) the deployment must use the LocalSandbox tool-execution backend, where tools run as local subprocesses that inherit the parent server's environment variables, rather than an isolated/remote sandbox (e.g. E2B) that does not pass through LETTA_PG_URI, POSTGRES_PASSWORD, and provider API keys. Impact is greatly amplified in MULTI-TENANT deployments where a single shared Postgres holds multiple tenants' data, since the leaked LETTA_PG_URI/POSTGRES_PASSWORD then grants cross-tenant access. Limiting factors: single-tenant deployments, instances running tools in a properly isolated sandbox, or configurations that scrub secrets from the tool environment are not exposed to cross-tenant data theft; exploitation also depends on the attacker already holding valid tool-execution credentials. |
| Impact | This is a moderate-priority confidentiality flaw whose real-world risk hinges on deployment model rather than raw reachability. The CVSS 4.0 vector (AV:N/AC:L/AT:N/PR:L/UI:N) and the 7.1 score indicate a network-reachable, low-complexity attack that nonetheless REQUIRES authentication (PR:L) - exploitation is not anonymous. Impact is confined to confidentiality (VC:H, with VI:N/VA:N), consistent with CWE-526 (sensitive information exposed via environment variables): the attacker reads secrets, they cannot directly tamper or destroy. The vector reports SC:N (no subsequent-system impact), yet the description explicitly states that stolen DB credentials enable 'cross-tenant database data' access - this is a notable TENSION: the scored vector arguably under-weights the downstream multi-tenant breach, so the effective business impact is higher than the base metrics suggest. Public exploit code exists (reported by VulnCheck and Lasso Security) which raises the practical likelihood of use, but there is NO CISA KEV listing and NO EPSS score provided, so there is no evidence of widespread automated exploitation - treat it as a credible, documented technique rather than mass-exploited. Net: prioritize for any multi-tenant Letta deployment running the LocalSandbox, deprioritize for single-tenant or fully-sandboxed setups. |
| Exploit ScenarioDerived | An authenticated low-privilege user on a shared, multi-tenant Letta instance submits code through POST /v1/tools/run or registers a custom tool; because the LocalSandbox subprocess inherits the full server environment, the code simply reads environment variables to harvest LETTA_PG_URI, POSTGRES_PASSWORD, and provider API keys. The attacker then uses the leaked Postgres connection string to query the shared database directly and read other tenants' data. Public proof-of-concept/exploit material is available (VulnCheck/Lasso Security), lowering the effort to reproduce this network-reachable, low-complexity attack. |
Get an e-mail when this CVE is exploited, patched or re-scored. Free account.
Alert meAnalysisAIv1 – Oct 20, 2026
Authenticated users of Letta through 0.16.8 can read server secrets when the LocalSandbox tool backend is used, because tool subprocesses inherit the server process environment; calling POST /v1/tools/run or running a custom tool can expose LETTA_PG_URI, POSTGRES_PASSWORD, and provider API keys. In multi-tenant deployments sharing one Postgres instance, those leaked database credentials can be used to reach other tenants' data, although single-tenant or properly isolated sandbox deployments are much less exposed. Publicly available exploit code exists, but the provided data does not confirm active exploitation (no CISA KEV entry), and exploitation requires an authenticated account with permission to execute tools.
RemediationAI
The provided data does not identify a vendor-released patched version, so no exact fixed release can be cited; versions through 0.16.8 are affected. Until a fix is available, stop using LocalSandbox for untrusted or lower-privileged tool execution and move to an isolated or remote sandbox backend that does not inherit the server environment, such as E2B-style isolation, accepting added latency and an external sandbox dependency. If LocalSandbox must remain, scrub LETTA_PG_URI, POSTGRES_PASSWORD, and provider API keys from the subprocess environment before tool launch, or run tools under a dedicated low-privilege OS account with a minimal environment, accepting that tools needing those credentials may break. Use per-tenant or least-privilege database roles so a leaked LETTA_PG_URI or POSTGRES_PASSWORD cannot read other tenants' data; this may require schema and connection changes. Restrict POST /v1/tools/run and custom tool creation to trusted administrators, monitor for environment-dumping commands, and rotate any possibly exposed database and provider secrets. Advisory URLs: https://www.vulncheck.com/advisories/letta-through-0.16.8-credential-exposure-via-tool-sandbox-environment-inheritance and https://www.lasso.security/vulnerability-database/cve-2026-77938; source: https://github.com/letta-ai/letta/blob/1131535716e8a31c9a437f8695e25ac98f203a24/letta/services/tool_sandbox/base.py#L487.
Affected VersionsENISA
Affected Packages
Source: OSV (Open Source Vulnerabilities)
Affects your stack
Affects 1 package in euvd
Checking this CVE against your own stacks and servers is part of the Basic plan and up.
Technical ContextAI · show details
Letta is an AI agent and tool-execution platform; the flaw is CWE-526, exposure of sensitive information through environmental variables. In the LocalSandbox implementation referenced in the vendor source, tools are launched as subprocesses that inherit the parent server's full environment instead of a scrubbed allowlist. Letta's tool execution API (POST /v1/tools/run) and custom tool registration provide the execution path, so an authenticated user with tool-run privileges can print environment variables and capture LETTA_PG_URI, POSTGRES_PASSWORD, and model/provider API keys. The CVSS 4.0 vector AV:N/AC:L/PR:L/UI:N/VC:H/VI:N/VA:N and our assessed CVSS 3.1 AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:N describe a confidentiality-only, authenticated issue. Impact is deployment-dependent: LocalSandbox plus a shared multi-tenant Postgres raises it to cross-tenant data access; isolated remote sandboxes or environment-scrubbed configurations do not expose these secrets. CPE cpe:2.3:a:letta-ai:letta:*:*:*:*:*:*:*:* covers Letta, with the description scoping affected versions through 0.16.8.
Know when Letta gets its next vulnerability
- Watch products and get an email when something new is published
- Check this CVE against your own stack and servers
- Exploit and PoC references, threat intelligence from 22 sources
No credit card · 7-day full trial
Severity by source
CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:H/VI:N/VA:N/SC:N/SI:N/SA:N
Network API reachable (AV:N) with low complexity (AC:L) but needs an authenticated tool-running account (PR:L); confidentiality-only secret disclosure (C:H, I:N/A:N), scope unchanged to align with the input's SC:N.
Primary rating from Vendor (VulnCheck).
CVSS VectorVendor: VulnCheck
DescriptionCVE.org
Letta through 0.16.8 contains an information disclosure vulnerability that allows users running tools to read server secrets because LocalSandbox subprocesses inherit the full server environment. Attackers can run code via POST /v1/tools/run or custom tools to read LETTA_PG_URI, POSTGRES_PASSWORD, and provider API keys, then access cross-tenant database data.
Technical POC Analysis
CWE-526, Exposure of Sensitive Information Through Environmental Variables.
Full technical POC analysis - root cause, step-by-step exploitation and patch breakdown - is part of the Pro and Business plans.
Executive ViewAI
Letta through 0.16.8 can leak its own server-side secrets - the Postgres connection string and password plus third-party model provider API keys - to any authenticated user who is allowed to execute tools, because tool subprocesses inherit the server process environment. Leadership should care because a single ordinary tool-enabled account can walk away with credentials that, in multi-tenant deployments sharing one Postgres instance, open the door to other customers' data and to billable provider accounts. Publicly available exploit code exists, but the supplied data does not confirm active exploitation.
No vendor-released patch identified at time of analysis. Interim mitigations: disable the LocalSandbox tool backend or replace it with an isolated, non-inheriting sandbox; limit tool-execution permission to a small set of trusted accounts; rotate every secret exposed to the server environment (LETTA_PG_URI, POSTGRES_PASSWORD, provider API keys); issue per-tenant Postgres credentials or separate instances; segment network egress from the Letta host to database endpoints; and monitor POST /v1/tools/run plus custom tool activity for anomalous secret harvesting.
Threat Intelligence Overview
How this decision was reached
The engine reads each signal below and applies a fixed list of rules from top to bottom. The first rule that matches sets the recommended action – like a statement, every line is shown.
What every action level means
SSVC AssessmentENISA
Priority Score
Attack Technique
Information disclosure occurs when an application unintentionally exposes sensitive data that aids attackers in reconnaissance or directly compromises security.
Related Vulnerabilities in Letta Authentication and Information Disclosure Flaws
View all 3Improper neutralization of directives in dynamically evaluated code within letta-ai letta 0.16.4 allows remote attackers
Server-side request forgery in letta-ai letta 0.16.4 allows authenticated remote attackers to manipulate ImageContent pa
EUVD-2026-104694
GHSA-jxmg-qw9j-hxj9
NVD
CVE.org
EPSS
Vulnrichment
Exploit-DB