CVE-2026-76072
HighCVSS 7.4Exploitation Probability (EPSS)
Low risk24th percentile - higher than 24% of all known CVEs
Summary
The Continue CLI in unattended mode (headless/auto) uses an incomplete denylist as its only barrier to destructive shell commands. It allows the Bash tool and blocks only certain paths, permitting recursive forced removal of directories like /home, /root, /var, /opt, or /srv. Additionally, due to variable parsing, the command rm -rf $HOME is not blocked, and other dangerous tools (find, shred, wipefs, truncate, pkexec) are not properly handled.
Risk Assessment
Malicious content (e.g., a web page, repository file, or issue text) can cause the unattended run to delete user data, leading to data loss or system integrity compromise.
Recommendation
Immediately update the Continue CLI to a patched version that expands the list of blocked commands and paths. Until then, avoid running in unattended mode on data whose loss would be critical.
Related vulnerabilities
- CVE-2026-82688Critical
D-Link DNS-340L and DNS-345 (versions 1.01B04/1.03B06/1.04.B02/1.05b04) have a vulnerability in /cgi-bin/virtual_vol.cgi (Virtual Volume Handler). Manipulation of arguments f_sharename/f_target/f_name leads to OS command injection. Remote exploitation is possible, and the exploit has been publicly disclosed.
- CVE-2026-49003Critical
Attackers can exploit command injection vulnerabilities to delete core system runtime files, causing the monitoring module to crash. They can also gain root privileges to steal configuration passwords (e.g., SNMP) and tamper with critical system parameters, leading to abnormal operation of the entire power system.
- CVE-2026-82874Critical
ToolJet before v3.16.208 fails to validate that authenticated users belong to the organization specified in the organizationId path parameter of tooljet-db endpoints, allowing any Builder user to read, modify, and delete tables across tenant boundaries. Attackers can extract victim organization IDs from public app endpoints, then exploit schema operation endpoints to disclose schemas, plant malicious tables, corrupt schemas, or permanently destroy data.
- CVE-2026-82872Critical
ToolJet before v3.16.208 fails to validate that the path organizationId matches the authenticated user's workspace before performing ToolJet DB table operations. A workspace admin can create, view, and delete database tables in another workspace by replacing the organizationId parameter in table-management API requests.
- CVE-2026-82870Critical
ToolJet before v3.16.208 fails to validate organizationId ownership in database write and destroy routes, allowing any builder-role user to create, alter, or drop tables in other organizations' databases. Attackers can exploit missing guards to permanently delete tables, insert arbitrary data, and modify schemas across tenant boundaries on shared instances.
- CVE-2026-82860Critical
@hulumi/policies before 1.3.2 fail to fully inspect inline and attached IAM policy evidence for the administrator-policy guardrail. Attackers can craft admin-equivalent policy paths that bypass policy evaluation controls.
- CVE-2026-82859Critical
hulumi before v1.3.2 contain a deployment SCP template that allows tag-on-create bypasses for hulumi:iac-role protections. Attackers can bypass intended IAM boundary restrictions by exploiting the weakened SCP template in downstream deployments.
- CVE-2026-82858Critical
@hulumi/drift before 1.3.2 accept externally supplied execute plans without sufficient provenance validation, allowing untrusted reconciliation input to be treated as trusted. Attackers can supply malicious execute plans that bypass security checks to perform unsafe reconciliation operations.
- CVE-2026-82857Critical
A privilege escalation vulnerability in hulumi before v1.3.2 exists in the weekly integration IAM policy, allowing role lifecycle operations on af-e2e-* roles without sufficient boundary restrictions. Attackers with the documented principal can create persistent higher-privilege roles in the sandbox account.
- CVE-2026-82856Critical
A vulnerability in @hulumi/policies before 1.3.2 fails to properly validate set-qualified AWS IAM condition operators in GitHub OIDC trust policies. Attackers can use ForAnyValue:StringLike operators to hide wildcard GitHub Actions OIDC subject conditions from security guardrails.
Original NVD description (English source)
The Continue CLI applies an incomplete denylist as its only barrier to destructive shell commands when running unattended. In headless mode and auto mode the default policy in extensions/cli/src/permissions/defaultPolicies.ts grants the Bash tool the allow permission, and permissionChecker.ts hard-blocks a command only when the terminal-security evaluator returns a disabled verdict, so isCriticalCommand in packages/terminal-security/src/evaluateTerminalCommandSecurity.ts is the sole control. Its dangerous-path test matches only /, /*, ~, ~/*, /usr, /etc, /bin and /sbin and their prefixes, so a recursive forced removal of /home, /root, /var, /opt or /srv is not disabled. The command line is parsed with shell-quote, which reduces $HOME to an empty token, so rm -rf $HOME also fails the dangerous-path test while the shell re-expands the variable when the command is spawned. find with -delete is rated high risk rather than disabled, and shred, wipefs, truncate and pkexec are not handled. Because the agent autonomously reads content it does not control, including fetched web pages, repository files and issue text, an indirect prompt injection in that content can cause an unattended run to destroy the invoking user's data.