How it works
- The MX sends two kinds of syslog:
urls(each HTTP request, and each HTTPS session with its host name) andflows(each connection that crosses the firewall, with IPs and ports). With the URLs role it also sends the content filtering blocks (content_filtering_block). - The SecureAI collector, a container that runs in your network, takes the host out of each
urlsline, keeps only the ones that go to an AI service and sends them to your tenant over HTTPS, without the query string. The rest of the browsing never leaves your network. - SecureAI reads the client’s IP and MAC and the destination host from each line, and shows it under Inventory →
Network Sources with the
cisco-merakiconnector. - When you turn on blocking, SecureAI writes its list into the Blocked URL patterns of each network you picked. It syncs again a few seconds after each block or release, and every 5 minutes.
Requirements
- An MX with the Advanced Security license. Without it, the network has no content filtering and SecureAI has nowhere to write the block. Discovery works anyway.
- A Dashboard API key from an administrator with write (full) access to the networks where SecureAI will block. The API has to be enabled for the organization (Organization → Settings → Dashboard API access). You generate the key under My profile → API access.
- A Linux machine with Docker on the MX’s network for the collector. The MX must reach it over syslog, and it must reach your tenant over HTTPS.
- A SecureAI administrator with a second factor set up.
1. Connect the MX in SecureAI
- Open Admin → Integrations and pick the Network / SWG category.
- On the Cisco Meraki MX card, click Connect.
- Under Add a Cisco Meraki MX, enter a name that lets you recognize the device and, if you want, the collector IP.
- Click Create and show the configuration. SecureAI asks for your second factor, creates the log source and the
feed the collector uses to filter, offers Download the collector (.zip) and shows
collector.envand the Values for the Dashboard.
2. Install the collector
- Copy the
.zipto a Linux machine with Docker on the MX’s network and unzip it. - In the
secureai-collectorfolder, rundocker compose up -d.
3. Send the MX syslog to the collector
In the Dashboard: Network-wide → General, Reporting section, Add a syslog server:- Server IP: the collector IP.
- Port: 514.
- Roles: URLs and Flows.
firewall instead of flows;
SecureAI reads both.
4. Turn on blocking
In the Cisco Meraki MX panel, Blocking tab:- Paste the Dashboard API key.
- Show my MX networks lists the networks with an MX in the organizations that key sees. Pick the networks. If
the key can’t list organizations, type the network ids (
L_…orN_…). - Check I understand that SecureAI writes into the blocked URL patterns of these networks.
- Click Turn on blocking. SecureAI asks for your second factor and runs the first sync.
5. Check
- In the Dashboard, Security & SD-WAN → Content filtering of a network you picked: Blocked URL patterns shows the domains of the SecureAI block list, after your own.
- From a machine behind the MX, open an AI service you prohibited. According to Cisco’s documentation, the MX shows its block page over HTTP and drops the connection over HTTPS.
- Under Inventory → Network Sources, the IP of that machine shows up with the
cisco-merakiconnector.
Turn off blocking
Turn off, under Response & auto-block, asks for your second factor. SecureAI removes the patterns it wrote from each network and leaves yours, and the MX stops blocking those domains. The Activity Log records it with critical severity. If a network doesn’t accept the change, blocking stays on and that network keeps the SecureAI patterns until you turn it off again. If you pause network enforcement in SecureAI, SecureAI stops writing to the networks and the MX keeps blocking the last list.What SecureAI does when something fails
The panel shows the code of the last failure.
Limits
- Without decryption, only the host. The MX doesn’t decrypt HTTPS: the
urlslog of an HTTPS session carries the host and the methodUNKNOWN, with no path. Content filtering also decides by host, so SecureAI blocks whole domains. flowslines don’t add to the inventory. They carry IPs and ports, no host: SecureAI accepts them and doesn’t count them.- Content filtering blocks may not be counted. In Cisco’s documentation example, the
content_filtering_blocklog carries the client’s MAC and not its IP, and SecureAI can’t tie it to a source. If your firmware addsclient_ip, SecureAI counts it as blocked. - Entry limit. Neither the API reference
(Update Network Appliance Content Filtering)
nor the MX content filtering guide publishes a maximum number of blocked patterns, so SecureAI doesn’t cut by
default. On a test network with no device, the Dashboard accepted and saved 20,000 patterns; we don’t know how many
an MX applies. If the Dashboard rejects the list, that network keeps the one it had and the panel shows
MERAKI_BAD_REQUEST. If you know your organization’s limit, ask support to set it: above it, SecureAI doesn’t write to any network and warns you withMERAKI_LIMIT. - Wildcards. The MX matches by stripping subdomains from left to right, so
openai.comcoversapi.openai.com, and an asterisk inside a URL counts as text (How to Filter Web Content Using MX and Cisco Talos). That’s why SecureAI writes*.openai.comasopenai.com. A wildcard in the middle (bedrock.*.amazonaws.com) can’t be written without widening it, and SecureAI leaves it out; the panel says how many patterns it left out. - A block reaches the whole network behind the MX, not a user or an IP.
- Meraki has a “Generative AI” category. SecureAI doesn’t use it: it blocks by domain, provider by provider, so what you approve keeps working. If you turn it on by hand in the Dashboard, it can also block the providers you approved in SecureAI.

