DNS

DNS tunneling is a classic covert data exfiltration technique that encodes data into DNS query requests; because the DNS protocol must be allowed through in almost all network environments, traditional firewalls find it very difficult to block completely.

An A record is used to map a domain name directly to an IPv4 address, while a CNAME record is used to point one domain name to another domain name.

DNS hijacking refers to attackers tampering with DNS resolution results to redirect users' access to legitimate websites toward malicious servers

DNS queries are almost never blocked by enterprise firewalls, and traditional security devices focus their monitoring on HTTP/HTTPS traffic while giving insufficient scrutiny to DNS traffic.

The TTL (Time To Live) value determines how long DNS resolution results survive in the cache and directly affects the speed at which DNS changes take effect.

Users can block their TV from sending telemetry data to manufacturer servers through router or DNS-level tools.

The hierarchical structure of the domain name system is: ICANN (top-level governance) → registries (managing specific top-level domains such as .com and .net) → registrars (providing registration services to end users).

DNS propagation typically requires anywhere from several hours to 48 hours.

This hijacking experiment redirects the PS5's RTMP connection to a self-built RTMP receiving server by controlling DNS resolution or routing on the local network.

When enterprises deploy AI agents, they should incorporate DNS anomaly detection into security monitoring to identify DNS tunnel fingerprints such as abnormally long subdomains, high-frequency resolution requests, and unconventional domain names.

Even when a sandbox blocks conventional outbound HTTP/HTTPS connections, DNS resolution requests are often still permitted, thereby creating an escape path.

Nearly all network environments allow DNS query traffic to pass through, making it susceptible to abuse as a channel for data exfiltration or command delivery.

To address the DNS tunneling risk for AI agents, recommended protective measures include DNS egress monitoring and filtering, enforcing the use of controlled DNS resolvers, least-privilege network design (deny outbound traffic by default), and behavioral auditing.

A single DNS query can carry approximately 100–200 bytes of data, but a continuous data stream can be pieced together through high-frequency queries.

DNS defaults to UDP port 53, which is rarely included in outbound blocking rules.

Even when a container's default gateway is pointed to an internal DNS resolver, an agent that can directly construct UDP packets (via raw sockets or low-level network APIs) may still be able to bypass system-level DNS settings and connect directly to external authoritative servers.

Configuring a dedicated domain for each AI Agent is extremely costly, requiring individual setup of DNS, SPF, DKIM, and DMARC along with a lengthy IP warm-up process.

Domain ownership verification is typically achieved by requiring enterprises to add a specific TXT record to their DNS records or place a verification file in the website's root directory.

Switching to a public DNS service (such as Cloudflare's 1.1.1.1 or Google's 8.8.8.8) can bypass an ISP's DNS-layer blocking, because these servers are not subject to Spanish regulatory directives.

Users can bypass domain-level or IP-level blocking by using a VPN, changing their DNS settings (such as using a public DNS service), or accessing directly via IP address.