Visibility → Enforcement → Proof
Your AI is in production.
Do you know what it’s doing?
Agents are deploying faster than teams can monitor them. GLACIS gives you behavioral benchmarks in minutes, runtime enforcement when you’re ready, and cryptographic proof that your controls actually ran.
Free, open source. pip install glacis-autoredteam
The journey
Visibility first.
Enforcement second.
Proof third.
Start with a free scan of any AI system. When you’re ready, add runtime enforcement. Proof accumulates automatically while you work.
auto-redteam
Free · Open Source
Point at any AI system. Get behavioral benchmarks in minutes. Toxicity, hallucination, jailbreak resistance, PII leakage, prompt injection — all version-pinned and repeatable.
pip install glacis-autoredteam
arbiter
SaaS Subscription
Sidecar runtime + cloud control plane. Continuous enforcement of governance policies across your entire AI fleet. Drift detection, shadow-to-enforce mode, real-time dashboards.
- Fleet-wide visibility
- Permit / deny / escalate
- Purpose-built SLM (zero egress)
witness network
Enterprise
Cryptographic receipts for every consequential AI decision. Third-party witnessed, tamper-evident, independently verifiable. Compliance for free — proof accumulates while you use the other tools.
- Tamper-proof audit trail
- Independent verification
- Zero-egress architecture
Featured in
How it works
Nobody else spans visibility → enforcement → proof
Eval tools stop at benchmarks. GRC platforms stop at dashboards. GLACIS connects the entire journey — from your first scan to a cryptographic receipt that stands up in regulatory review.
One tool, three stages
Scan your AI today. Add enforcement when you’re ready. Evidence accumulates automatically. No manual stitching, no separate tools, no context loss.
Purpose-built SLM
Our specialized small language model runs locally, producing version-pinned scores. Your prompts, responses, and PHI never leave your environment.
Evidence that compounds
Every interaction generates a cryptographic receipt. Over time, you build an auditable evidence chain — so when compliance becomes a requirement, your posture is already built.
Beyond dashboards
Eval tools give you a snapshot. Compliance platforms give you a checkbox. GLACIS gives you independently verifiable, per-inference evidence that your controls actually ran.
Zero Egress
Data stays local
Shadow → Enforce
Go live when ready
Tamper-proof
Crypto signatures
<50ms
Total overhead (p95)
Why now
Agentic sprawl meets regulatory convergence
AI agents are deploying at a pace that outstrips every governance mechanism built for the last generation of software. The regulatory response is accelerating at the same speed.
The OpenClaw moment
OpenClaw hit 247k GitHub stars and 12% of its marketplace was immediately compromised. Agent ecosystems are growing faster than anyone can vet them. If you can’t see what your AI is doing, you can’t secure it.
Every compromised agent tool is a potential supply chain attack on every system that calls it. Visibility isn’t optional — it’s the prerequisite.
Three deadlines, one summer
Colorado AI Act — June 2026
Rebuttable presumption of reasonable care for deployers who demonstrate NIST AI RMF compliance with verifiable evidence.
EU AI Act — August 2026
High-risk AI obligations go live. Articles 12 and 14 require logging, human oversight, and verifiable records.
California ADMT — Already in effect
Automated decision-making technology disclosures and opt-out requirements.
The window is closing.
Organizations that build visibility now will have enforcement and proof by the time deadlines hit. Organizations that wait will be scrambling.
See it in action
Watch Governance Enforced in Real Time
An AI request arrives...
“We would never accept this for any other critical system. Financial systems have audit trails. Medical devices have mandated records. Aircraft have flight recorders. AI systems need the same level of verifiable evidence.”
Who we help
Visibility and proof for every stakeholder in your AI pipeline
Healthcare AI Vendors
Get through procurement faster with proof
Health Systems
Evidence for discovery — before you need it
Financial Services
SR 11-7 model risk with cryptographic proof
EU AI Act
Article 12 & 14 evidence, continuously proved
FAQ
Common questions
What is auto-redteam?
It’s our free, open-source behavioral benchmarking tool. Point it at any AI endpoint and get repeatable scores for toxicity, hallucination, prompt injection, PII leakage, and more. Version-pinned, so your results are comparable across runs and models.
We already have SOC 2 / are working toward HITRUST
Great — those cover IT controls. AI-specific assurance addresses model behavior, decision audit trails, and content safety risks that SOC 2 and HITRUST don’t cover. They’re complementary.
How is this different from Promptfoo or other eval tools?
Eval tools give you a snapshot. GLACIS gives you the full journey: behavioral benchmarks (auto-redteam) feed into runtime enforcement (Arbiter), which generates cryptographic proof (OVERT). Nobody else connects visibility to enforcement to independently verifiable evidence.
Does GLACIS just monitor, or does it actually enforce?
Both. You define controls declaratively — which guardrails to enforce, at what confidence thresholds, with what failure modes. The Arbiter evaluates every AI request against your active policy and makes real-time permit/deny decisions. You start in shadow mode (observe only) and transition to enforcement when ready. The transition itself is attested.
What if we’re not ready for enforcement yet?
Start with auto-redteam — it’s free and gives you immediate visibility. When you’re ready to enforce, add Arbiter. When you need proof for auditors or regulators, turn on OVERT. Each layer works independently, but they’re stronger together.
Start seeing what your AI is doing.
pip install glacis-autoredteam is the fastest way to behavioral benchmarks. Book a demo for runtime enforcement. Or take the free governance assessment.
pip install glacis-autoredteam
Behavioral benchmarks for any AI system. Runs locally. Results in minutes.
auto-redteam.com →Book a Demo
30-minute walkthrough of Arbiter runtime enforcement and OVERT attestation.
Schedule →Governance Assessment
Map your AI risk posture and surface gaps before deadlines hit.
Take Assessment →