Aegis
Governed AI assurance and adversarial validation.
Agencies and defense programs are fielding AI models and agents faster than they can produce evidence that those systems are safe to authorize. Aegis tests the safeguards around an AI system before it goes live, and turns every result into signed, machine-readable evidence that an Authorizing Official can check independently, including inside a disconnected enclave.
To save this brief as a PDF, use your browser print command (Ctrl+P on Windows, Cmd+P on macOS) and choose Save as PDF.
AI ships ahead of assurance.
Prompt injection, leakage of sensitive data and system prompts, unsafe handling of model output, and over-privileged agents are new attack surfaces that scanners and periodic penetration tests do not cover. AI security tools test the model but do not produce authorization-grade evidence. Governance tools map controls but do not test. Red team results arrive as reports and screenshots that no one can re-verify a year later.
The timing is set by policy. OMB M-25-21 requires minimum risk management practices, including pre-deployment testing, for high-impact AI, and M-25-22 carries testing expectations into AI acquisition. FedRAMP is moving to machine-readable OSCAL packages, DoD is moving toward continuous authorization, and AI is reaching disconnected and classified networks where cloud-only tools cannot follow.
- AI fielded with guardrails no one independently tested
- Agents with more reach than their mission requires
- Analyst hours rebuilding evidence for every review
- Findings that never reach a control owner
- Authorization delays and rework when an assessor rejects artifacts that cannot be verified
Governed testing in, verifiable evidence out.
What the decision makers get.
- Evidence packs verified on your own machine, with no network and without the vendor present
- OSCAL assessment results and a POA&M for the system of record, not slides and spreadsheets
- A record of who approved each test and each change, in an append-only, tamper-evident ledger
- Artifacts that regenerate to the same digests, so they can be re-checked a year later
- Pre-deployment guardrail test evidence for high-impact AI, in support of OMB M-25-21 practices
- Results mapped to the OWASP Top 10 for LLM Applications and MITRE ATLAS
- Agent containment findings before agents are fielded
- A consistent, benign test protocol to apply when acquiring AI under M-25-22
Where it is different.
- Evidence an assessor can verify offline. A standalone verifier rejects tampered, incomplete, unsigned, and forged packs and checks a legitimate pack with only the public key.
- One step from validation to OSCAL. Assessment results and POA&M come from the same control results that go into the signed pack.
- Guardrail testing without weaponized content. Benign canaries, run only inside approved scope and behind an allowlisted egress guard.
- Agent containment tied to real topology. Reach is computed from the actual dependency graph.
- Built for disconnected and high-side enclaves, with signed update bundles and no dependency on a hosted model or cloud console.
- Determinism as a compliance feature. The same inputs give the same digests, so artifacts can be regenerated and matched.
| Capability | Aegis | AI security and AI red-team tools | Breach and attack simulation | Federal GRC and OSCAL tools | Traditional penetration testing |
|---|---|---|---|---|---|
| AI guardrail probing | Yes, starter coverage | Yes, typically broad | Emerging | No | Varies by team |
| Agent containment tied to topology | Yes | Partial | No | No | No |
| OSCAL results, POA&M, eMASS-aligned export | Yes | No | Rare | Yes | No |
| Signed, offline-verifiable evidence | Yes | No | No | Partial | No |
| Disconnected and high-side profile | Yes | Rare | Rare | Rare | Manual |
| Runtime guardrail or AI firewall | No, by design | Yes | No | No | No |
Pilot on one AI system, then expand by module.
Scope
Agree the AI system or enclave, rules of engagement, approvers, and the controls the evidence should support.
Pilot
Run a governed guardrail and containment campaign and deliver a signed pack with OSCAL output your assessor verifies.
Expand
Add systems under test and the AI Purple Team module, so findings become detection rules the SOC validates.
Operationalize
Add the Federal Evidence module for recurring OSCAL delivery and extend to disconnected and high-side enclaves.
Packaging: a core platform with AI Red Team, AI Purple Team, and Federal Evidence modules. Pricing is scoped per program and per AI system under test.
Start with one AI system.
We run a sanitized campaign end to end and hand you the signed evidence pack and its OSCAL output. You verify it on a disconnected machine.
Test one AI system, then verify the pack yourself.
Sanitized demonstrations are available for classified environments.