Scanning policy
Sonar observes public infrastructure submitted by authenticated customers. We do not operate a global internet scanning index.
ASN routes are passive observations and do not establish authorization. A discovered range stays out of scan scope until a workspace submits written authorization and the operator reviews it. After approval, a workspace administrator can explicitly activate a range of at most 16 public addresses. Each generated IP must also pass its hosted verification challenge. Activation does not start a scan; ordinary workspace scan limits and controls still apply. Larger prefixes remain passive.
Authorization and scope
Domains require a DNS TXT verification record. Public IP addresses require a hosted HTTP verification challenge. Sonar checks the challenge before each scan. Domain ownership permits discovery within that domain; it does not establish permission to scan unrelated infrastructure or every tenant on a shared address. Customers remain responsible for the scope of their authorization.
Current scan behavior
Scans use DNS queries, public certificate-transparency lookups, TCP connection probes, HTTP requests and TLS handshakes. Port coverage is selectable per workspace from Tohou’s own curated list of commonly exposed TCP service ports (1,504, or a 228-port core profile), using TCP connection (not raw) probes at no more than 6 probes per second with one retry, swept in groups of up to 200 ports. Workspaces on the Pro plan can choose ports 1–20,000 plus the curated ports above 20,000, and workspaces on the Business plan all 65,535, swept the same way at no more than 25 probes per second. High-value services that fall outside common frequency lists — such as databases, message queues and container or orchestration control planes — are always included. Only for ports already found open, Sonar then runs low-intensity (level 2 of 9) service and version fingerprinting, and on deployments with the necessary privileges it also sends a single UDP probe, without retries, to DNS (53), NTP (123) and SNMP (161). When an address accepts connections on ports no service uses (a firewall or proxy answering for it), its port list is treated as unreliable and port-based findings are withheld.
Sonar correlates the software it fingerprints on a host to published vulnerabilities (CVEs): fingerprinted server software, service and operating-system components, and versioned client-side libraries are matched by version against upstream vulnerability databases. These findings assert that an affected version is exposed, not that the host is exploitable; each is labelled accordingly and prioritized using known-exploited status and exploit-probability scores. No exploit is attempted during this correlation.
Content-delivery networks and shared infrastructure
Before scanning, Sonar checks whether a host resolves to a known CDN, reverse-proxy or WAF edge (for example Cloudflare, Akamai, Fastly, Amazon CloudFront and others), by both published IP ranges and CNAME signatures. When a host sits behind such a provider, Sonar does not port-scan the shared edge address, which is infrastructure the customer does not own. Instead it confirms only the web ports and tests the website through its hostname (HTTP, TLS and header checks) exactly as a browser would. Where a verified domain is protected by a CDN but one of its own hosts resolves directly to an origin server, Sonar flags that host as a CDN bypass, because traffic to it skips the provider’s protection, and scans it as a customer-owned origin.
Sonar additionally performs read-only HTTP GET requests for a fixed list of about 60 signatures: commonly misconfigured paths (for example /.env, /.git/config, server-status and phpinfo pages, robots.txt, sitemap.xml and /.well-known/security.txt), exposed source-control and deployment files, debug endpoints, API specifications, and login or admin panels. The signature engine requests each of its paths at most once per web origin. The signatures run on the standard web ports and on up to four other open web ports per host (for example 8080 or 8443); the other path checks listed here run on the standard web ports only. Sonar also fetches the site favicon for application fingerprinting and captures a single page screenshot of a responding web surface. It enumerates subdomains from public certificate transparency, public archives and a common-name list, and resolves each candidate. No content is modified, no forms are submitted and no files are uploaded.
For a verified domain, Sonar reads the public registration record (RDAP) to warn about expiry and a missing transfer lock, and checks whether lookalike spellings of the domain are registered. Lookalikes are checked with ordinary DNS queries only; Sonar sends no traffic to the lookalike domains themselves.
Optional authenticated checks are limited to exact HTTPS paths on verified assets and a customer-provided test credential. HEAD checks collect headers; explicit GET checks compare response status with and without authentication and discard the body after at most 16 KiB. Workspace administrators must confirm each GET path is read-only. GET endpoints can still have side effects if the customer misidentifies an endpoint, so these checks do not submit forms or follow redirects. The scanner does not perform password guessing, exploitation or denial-of-service testing.
Each workspace runs at most one scan at a time, and the service runs a small, operator-configured number of workspace scans concurrently. Within a scan, network probes to a host are sequential, individually paced and rate-limited. Probes never use decoy addresses or spoofed sources, so all traffic originates from Sonar's own addresses. Outbound HTTP requests present standard browser User-Agent strings rather than a dedicated scanner identifier; operators who need to attribute traffic should match Sonar's source addresses and contact the address below. HTTP responses have time and size limits. Redirects are recorded rather than followed to other targets. Private, loopback, reserved and metadata addresses are excluded.
Authorized intrusive testing (opt-in)
Separately from the observation profile above, Sonar offers an opt-in tier of bounded active application checks for customers who are authorized to actively test their own assets. It is off by default and gated three ways: the operator must enable it for the deployment under written hosting-provider approval, a workspace administrator must turn it on, and each asset requires its own explicit written consent. Only an asset meeting all three conditions is actively tested.
For a consented asset, Sonar first performs a shallow, same-origin crawl (read-only GET requests, capped pages) (and any published OpenAPI/Swagger spec) to map real endpoints and parameters, then runs detection-only, non-destructive active checks against them: a benign reflected-input marker (reflected XSS); a single error-trigger character and a repeatable time-delay payload (error-based and blind SQL injection); an external redirect marker (open redirect); a traversal payload reading a well-known system file (path traversal / local file inclusion); a harmless arithmetic expression (server-side template injection); a benign echo marker (OS command injection); and a link-local metadata URL (server-side request forgery to cloud metadata); plus CRLF injection, CORS and host-header checks, cookie flags, content-matched probes for commonly exposed files and consoles, GraphQL introspection and review of issued JWTs. Requests are limited to GET and read-like POST (search or lookup); logout, delete, checkout and similar paths, password and file fields and anti-forgery parameters are never requested or injected into, and no other HTTP methods are sent. Findings are recorded per endpoint and parameter, each class is capped, and total request volume has a hard budget. A customer may supply a test-account login recipe so the same checks run behind authentication; it is stored encrypted and used only on that asset. The checks never extract data beyond the single proof marker, modify state, perform denial-of-service or volume fuzzing, or carry out credential stuffing. Findings indicate a possible issue to confirm manually; they are not proof of exploitation. A workspace can withdraw consent or disable the tier at any time.
Related-asset discovery
Beyond scanning, Sonar looks for assets that may belong to a verified domain by reading public sources, DNS records and certificate data. What it finds is only ever a suggestion with its evidence; nothing outside the verified scope is scanned or contacted until the customer verifies it. Two optional checks reach third parties and are off by default: guessing cloud storage bucket names (existence-only HEAD requests to Amazon S3 and Google Cloud Storage and a DNS lookup for Azure) and following SPF, MX, NS and DMARC records into other domains. A workspace administrator or the operator must turn each on.
Inbound sensors and published telemetry
Sensors receive traffic; they do not send it. Receiving an unsolicited connection is not scanning, so the statements above still hold: we operate no global internet scanning index, and no Sonar probe uses a decoy or spoofed source.
A sensor is a host on a Tohou-owned address, dedicated to the purpose. It holds no customer data and serves no customer traffic. It accepts a connection, records the request and closes. It never responds with an exploit payload, never scans the sender, never fetches anything the sender references, and never tries to identify the person behind an address.
We record the source address, the time, the port and protocol, a bounded slice of the request, any matched signature, and the network and country the address belongs to. We publish the address, when it was first and last seen, rounded attempt counts, the ports targeted and any matched vulnerabilities. We never publish request payloads, any credential material, the addresses of the sensors themselves, or anything beyond the address that could identify a person. Attempted passwords are counted and discarded; none is ever sent to the platform or published, in any form, hashed or otherwise.
Entries carry a confidence tier, expire on their own and are removed once the activity stops. If an address you control appears and the problem is fixed, write to [email protected]. We ask for proof that you control the address, decide within thirty days and then stop publishing it — and we say plainly that the sensors keep recording, so renewed hostile traffic from the address will be recorded and published again. The privacy policy sets out the full route, the lawful basis and the retention periods.
Customer-run agents and cloud connectors
Internal network scanning runs from an agent the customer installs inside their own network. It connects out to Sonar, scans only private ranges an administrator declared (connect scans and passive banner reads, no exploit traffic or credential attempts) and refuses any non-private address. Cloud connectors use read-only provider credentials supplied by the customer and send no traffic to the customer's assets.
Customer controls
Customers can exclude individual assets, run manual scans, select a daily scan hour, disable scheduling, and cancel pending or running scans. Cancellation takes effect between bounded checks. Active jobs also stop when a workspace is suspended.
Report unexpected activity
Contact [email protected] with the target, source address, UTC timestamp and relevant request logs. Do not send credentials or unrelated personal information. The operator can suspend a workspace and stop its scans while investigating.
Hosting constraints
Scanning can be disabled globally. Deployment on a provider does not imply permission to scan; the operator must confirm applicable provider policies before enabling active observation.