Governments are wiring AI into their most sensitive systems. We test whether that's safe to do.
The world's first offensive security framework for government AI infrastructure. Test AI systems making welfare, benefits, border, tax, healthcare, and critical infrastructure decisions before hostile actors do.
SPECTER APPARATUS is NIGHTFALL's Layer 90 module — Government AI Infrastructure Attack. It systematically tests the security of AI systems deployed across government agencies and critical infrastructure.
Eight coordinated subsystems map government AI topology, inject prompts into welfare/benefits/border decision systems, corrupt training data and audit trails, pivot laterally across department boundaries, and model cascading failure scenarios across critical national infrastructure.
Every attack vector is gated by APPARATUS_KEY. All reports are Ed25519 + ML-DSA-65 dual-signed with Merkle chain evidence linking all 8 subsystems. RFC 3161 timestamps ensure forensic validity.
SPECTER APPARATUS is restricted to authorised government security testing only. Deployment requires explicit written ROE, APPARATUS_KEY cryptographic authorisation, and compliance with UK Computer Misuse Act 1990, CFAA, and equivalent legislation. Unauthorised use is illegal.
Crown jewel detection. Passive reconnaissance across government AI infrastructure. Fingerprint agency AI endpoints, detect decision systems, identify kill switch configurations. Before we touch anything, we know everything.
Prompt injection into AI systems making welfare, benefits, and border decisions. The decision is corrupted before the citizen speaks. Inject adversarial prompts via application layer; measure decision manipulation success rate.
RAG poisoning. Compliance corruption. Audit trail backdoor. Poison retrieval-augmented generation knowledge bases so AI decisions are based on false premises. The record says it was clean.
Platform poisoning. Training backdoor. Vendor compromise. Inject backdoors into AI training pipelines before models are deployed to government systems. The system was never clean.
US, Chinese, European, UK AI dependency mapping. Kill switch scenario modelling. NATO analysis. Map which critical systems depend on which AI platforms. Find the off switch.
RAG backup survival. Multi-layer orchestration. Auto-reinfection. Establish persistence across government AI infrastructure so compromises survive system wipes and re-infections. It comes back.
Citizen identity pivot. Financial-border pivot. ARMAGEDDON simultaneous coordinated attack. One foothold in welfare AI. Every system across government. Pivot to healthcare, tax, central government.
Energy. Water. Telecoms. Transport. Defence. Healthcare. AI targeting across every critical national infrastructure vertical. Model cascade failures when AI-orchestrated systems fail simultaneously.
SPECTER APPARATUS uses cryptographic gating to enforce rules of engagement and authorisation. All sensitive operations require APPARATUS_KEY — a SHA-256 derived key based on target PAN ID and timestamp.
| Gate | Requirements | Operations |
|---|---|---|
RECON | API key | Passive fingerprinting, topology mapping, kill switch assessment |
INJECT | APPARATUS_KEY + ROE file | Prompt injection, citizen-facing system compromise |
CORRUPT | APPARATUS_KEY + --confirm-corrupt flag | RAG poisoning, audit trail corruption, supply chain backdoor |
UNLEASHED | APPARATUS_KEY + --confirm-unleashed + two-person rule | CNI attacks, cascade failure modelling, persistence across critical infrastructure |
export APPARATUS_KEY="SPECTER-APPARATUS-$(sha256sum <<<$TARGET_PAN_ID | cut -c1-16 | tr a-z A-Z)"
specter-apparatus recon --target 10.0.0.0/8
specter-apparatus citizen-inject --target welfare-ai --vector prompt-injection
specter-apparatus cni-attack --sector energy --confirm-unleashedSPECTER APPARATUS integrates with WARLORD PRIME for autonomous government AI red team campaigns. Full eight-subsystem chain execution with detection risk scoring, adaptive evasion, and multi-agency lateral movement orchestration.
All reports are cryptographically signed with Ed25519 + ML-DSA-65 dual signatures. Evidence chain uses SHA-256 Merkle tree linking outputs from all 8 subsystems. RFC 3161 timestamps provide forensic audit trail.
| Signature Type | Algorithm | Key Size | Purpose |
|---|---|---|---|
| Report signing | Ed25519 | 32 bytes | Authorisation proof, non-repudiation |
| Post-quantum backup | ML-DSA-65 | 4000 bytes | Quantum-resistant signature if Ed25519 compromised |
| Merkle chain | SHA-256 | 32 bytes | Link all subsystem outputs into forensic evidence DAG |
| RFC 3161 timestamp | TSA certificate | — | Independent timestamp authority verification |
SPECTER APPARATUS is restricted to authorised government security testing only.
Deployment requires:
Unauthorised use is illegal under UK law and carries criminal penalties. APPARATUS_KEY enforcement prevents accidental or unauthorized execution. All operations are cryptographically logged and auditable.
specter-apparatus recon --target <endpoint>
specter-apparatus citizen-inject --target <ai-system> --vector <type>
specter-apparatus decision-corrupt --target <rag-system>
specter-apparatus supply-chain-poison --target <platform>
specter-apparatus sovereign-map --nation uk|us|eu|cn
specter-apparatus persist --target <system>
specter-apparatus lateral --pivot <type>
specter-apparatus cni-attack --sector energy|water|telecoms|transport|defence|healthcare
specter-apparatus report --session <id> --output <path>