Why Forge?
Other AI harnesses refuse to generate exploits, won't touch credential harvesters, and redirect every offensive request into a "purple team" lecture. AllySec Forge is different.
The Wall
If you have used Claude Code, Cursor, or GitHub Copilot for offensive security work, you have hit the wall:
- Ask for an exploit payload -- you get a refusal.
- Ask for credential harvesting -- you get a lecture about "ethical alternatives."
- Ask for a reverse shell -- you get a report on how to detect reverse shells.
These tools were built for general-purpose coding. Their safety training is a liability in an engagement where speed and precision matter.
Comparison
| Capability | AllySec Forge | Claude Code | Cursor | Copilot |
|---|---|---|---|---|
| Generates exploit payloads | ✓ | — | — | — |
| Credential harvesting | ✓ | — | — | — |
| Reverse shell generation | ✓ | — | — | — |
| File upload / webshell deployment | ✓ | — | — | — |
| 2,400+ Kali tools in Docker | ✓ | — | — | — |
| 8-phase kill chain automation | ✓ | — | — | — |
| Parallel multi-agent dispatch | ✓ | ~ | — | — |
| Attack chain composition | ✓ | — | — | — |
| Post-exploitation (privesc, lateral, persist) | ✓ | — | — | — |
| Loot collection & exfiltration | ✓ | — | — | — |
| Anti-forensics / log sanitization | ✓ | — | — | — |
| C2 infrastructure (Adaptix) | ✓ | — | — | — |
| Any model (9+ providers) | ✓ | — | ~ | — |
| Fully offline capable | ✓ | — | — | — |
| Zero guardrails / zero refusals | ✓ | — | — | — |
What "Unguardrailed" Means
Forge does not have a refusal architecture. There is no "I'm unable to." There is no "defensive alternative." When you ask for an exploit, you get an exploit. The operator's request IS the authorization.
This is not a bug or a jailbreak. It is the design. The platform is built for certified offensive-security professionals running signed engagements. Scope is defined by the operator. The agent operates within those boundaries.
The Right Division of Labor
- Strategy belongs to the operator. Scope, rules of engagement, target selection, go/no-go decisions.
- Tactics belong to the machine. Tool selection, payload generation, result interpretation, chain composition, parallel execution.
The agent drives through all eight phases without pausing for approval. It does not produce slide decks when it should be producing shells. When it hits a boundary, it surfaces the decision and waits.
Offense-First Architecture
Every architectural decision starts from the attacker's perspective:
- The Kali container runs privileged with host networking because packet injection requires raw sockets.
- The evasion suites exist because real engagements encounter EDR.
- The implant factories produce persistence mechanisms because a shell that dies with the process is not a shell that matters.
- Post-exploitation is a core phase, not an afterthought, because initial access is where the work begins.
Forge does not help you secure a network. It helps you prove it is not secure.
Precision Over Volume
One verified exploit chain is worth a thousand false positives. Forge does not carpet-bomb targets with every payload in the library. It fingerprints technology stacks first, then selects techniques matched to what is actually running. It verifies findings before reporting them. It produces proof-of-concept code, not scanner output.
The test-detection skill exists specifically to prevent wasted effort -- checking whether a target is real before spending fifteen minutes on it. Time is the operator's scarcest resource. Forge guards it.