Attack vectors decide what the agents may attempt against your scope. You pick them in the **Attack vectors** step of the profile form, and agents plan their attacks within your selection. This page lists every vector, what it does, and what it may disturb.

## Safe and Intrusive

Every vector carries a safety label:

- **Safe** vectors only observe. They don't sign in and don't change anything on the target.
- **Intrusive** vectors act on the target. They can lock accounts, fill logs, or slow a service down. Select them only with written authorization from the owner of every target in scope.

For a first run on production systems, select only the safe vectors. Add intrusive ones once you've agreed on a test window with the service owners.

## Vectors

| Vector | Safety | What the agent does |
| --- | --- | --- |
| Exposed Services | Safe | Finds services listening on the hosts in scope. It doesn't sign in. |
| Insecure Configuration | Safe | Reads exposed configuration. It doesn't change settings. |
| TLS Configuration | Safe | Checks certificates and protocol settings. It doesn't alter the service. |
| Remote Administrator Exposure | Safe | Looks for reachable remote admin interfaces. It doesn't sign in. |
| Default Credentials | Intrusive | Tries well-known default passwords. This can lock accounts. |
| Authentication Enumeration | Intrusive | Probes which accounts exist. This can fill logs. |
| Injections | Intrusive | Sends injection payloads. This can disturb the services in scope. |
| Rate-limit Weaknesses | Intrusive | Sends repeated requests. This can throttle the service. |
| Weak Credentials | Intrusive | Guesses or brute-forces credentials of discovered services. This can lock accounts. |
| Authentication Bypass | Intrusive | Attempts to bypass authentication. It doesn't brute-force credentials. |
| Unpatched Known Vulnerabilities | Intrusive | Tries to exploit the CVEs that match each service version, using public proofs of concept. This can disturb the services in scope. |

With **Unpatched Known Vulnerabilities** selected, agents look up the CVEs for every fingerprinted service version, search for a public proof of concept, and try to exploit each one. Only exploited CVEs are reported with their proof. Denial-of-service CVEs are reported without being run.

Without it, CVEs matching a service version are reported at **Info** severity, without testing.

## What Agents Look For

Within the vectors you allow, agents use checks built for network services:

- **Exposed services**: unauthenticated datastores, JDWP debug ports, Node.js inspector ports, exposed PHP-FPM, SNMP with default read or write communities, and Modbus devices that accept unauthenticated writes.
- **Credentials**: vendor-documented default credentials on services that authenticate, and SSH password testing paced to respect OpenSSH source penalties.
- **TLS**: SSLv3, TLS 1.0 and 1.1, RC4, 3DES (Sweet32), CBC ciphers on TLS 1.0 (BEAST), NULL and anonymous ciphers, small RSA keys, SHA-1 certificate signatures, and expired, self-signed, or name-mismatched certificates.
- **DNS**: zone transfers, open recursive resolvers, unauthenticated dynamic updates, DNS version disclosure, DNSSEC turned off, private IP addresses in public records, and sensitive data in TXT records.
- **Web surfaces**: basic checks on web servers found on open ports, including server configuration disclosure.

## Rules Every Agent Follows

Whatever vectors you select:

- Testing stays non-destructive.
- Port scans stay bounded.
- Denial-of-service CVEs are reported, never run.
- SQL and NoSQL injections are proven with read-only oracles. Agents never modify or delete your data.
- Every finding is reproduced by a separate reporter agent, with its own evidence, before it becomes an issue.

## Not Tested

AI Network Pentesting doesn't cover:

- Signed-in application testing: access control, business logic, and authenticated API flows. Use [AI Pentesting](<https://docs.escape.tech/documentation/ai-pentesting/index.md>).
- Active Directory.
- Phishing and social engineering.
- Asset discovery beyond the scope you enter, such as subdomain enumeration.

## Related

- [Guardrails](<https://docs.escape.tech/documentation/network-pentesting/guardrails/index.md>): rate limit, duration, and stopping an assessment
- [Scope](<https://docs.escape.tech/documentation/network-pentesting/scope/index.md>): which targets the vectors apply to
- [Results](<https://docs.escape.tech/documentation/network-pentesting/results/index.md>): how findings and their proof appear