GLACIS

GLACIS is a runtime control layer for AI actions that also produces independently checkable decision records. It is best suited to AI vendors serving regulated customers, healthcare oversight teams, and operators of consequential agent workflows. Its combination of intervention and open-format records is useful, but it does not replace broader safety evidence or remove the need for manual review preparation.

Overview

When an action crosses configured bounds, GLACIS can allow it unchanged, constrain it, hold it for a person, or deny it. That gives teams a way to intervene before a risky action proceeds instead of relying solely on post-event analysis. The signed record proves the integrity of that record, not that the action was correct, that every event was captured, or that the overall system is safe or compliant.

Key features

Runtime controls and deployment

Prompt-injection defense, PII redaction, and output guardrails address common control points in AI workflows. The hybrid deployment model supports API, web, and self-hosted use. GLACIS says its managed service operates nine production components while working with the identity provider, cloud, and security systems a customer already has. That suits organizations seeking managed infrastructure without replacing their existing stack.

Signed records and OVERT

Covered decisions are signed and use OVERT, a published open record format that implementers and reviewers can inspect without depending on a proprietary schema. Anyone can check a record in a browser without a GLACIS account, which makes independent review practical for customers, auditors, or partners. Payloads remain in the customer’s environment and records carry hashes; protected fields can be omitted, depending on deployment configuration.

Evidence and verification boundaries

GLACIS does not currently export evidence packs automatically. A team must assemble a pack manually for each review, so organizations with frequent audits should budget operational time around the product. Browser verification is valuable for record integrity, but it should not be presented as proof of decision quality, complete logging, system safety, or regulatory compliance.

Pricing

GLACIS uses a freemium model. The Free account costs free, requires no card, and takes about a minute to set up. A 14-day trial is also offered. The free account is a sensible way for evaluators and small teams to assess the control model, while larger managed deployments should expect custom pricing. The published terms do not provide seat or usage quotas, so buyers should confirm capacity and renewal conditions before committing beyond evaluation.

The contact process describes a six-week evaluation on the workflow under review. GLACIS typically responds to contact requests within one business day and accepts only a small number of managed deployments each quarter, factors that may matter when timing a production rollout.

Platforms

GLACIS supports API, self-hosted, and web access. Its hybrid deployment can accommodate teams that need local control over payloads while using managed components for operations.

Who it's for

The strongest fit is an AI seller serving healthcare or another regulated buyer, a healthcare AI overseer, or a team building or operating AI agents where human intervention and reviewable records matter. It also suits people evaluating governance approaches through the six-week workflow review. Teams needing fully automated audit packages, guaranteed event completeness, or a direct safety and compliance determination should look beyond GLACIS or pair it with other controls.

Pros and cons

Pros

• Runtime intervention: Actions can be constrained, paused for a person, denied, or passed through, giving operators a choice of response when bounds are crossed.
• Independent verification: Signed OVERT records can be checked in a browser without a GLACIS account.
• Open record format: OVERT is inspectable and usable without a proprietary schema.
• Environment-conscious handling: Payloads stay in the customer environment and records use hashes, with protected fields optionally omitted by deployment.
• Broad delivery options: API, self-hosted, web, and hybrid deployment cover varied operating models.

Cons

• Manual evidence work: Evidence packs must be assembled by hand for each review because automated export is not offered.
• Verification is deliberately narrow: A valid record does not show that an action was right, that logging was complete, or that the system is safe or compliant.
• Limited managed capacity: GLACIS takes on a small number of managed deployments each quarter, which may constrain scheduling.
• No bug bounty: The vulnerability policy sets response targets but offers no bounty program.

Alternatives

LLM Security Tools is the broader category starting point when you want to compare runtime-security products rather than focus on signed decision records.

Choose PromptGuard instead if you want a freemium scanner with extensions and desktop platforms: its free tier allows 20,000 scans per month, one API key, one project, and 24-hour log retention, while Team costs $19 a month.

Prompt Inspector is a better match when detection quotas, REST API, SDK, and MCP access are the priority; its free plan covers 1,000 detections per month.

Pick NVIDIA NeMo Guardrails for a free, open-source Python package when you are prepared to arrange any external models and services yourself.

Amazon Bedrock Guardrails suits teams already buying AWS controls and willing to pay usage-based fees for content filters and denied topics.

Choose Openlayer Guardrails when you need a freemium workspace centered on inference pipelines, tests, logs, and project retention.

LLM Guard is a free open-source option, but its repository is archived and no longer actively maintained.

CyberSecEval is appropriate when an open-source benchmark suite, rather than runtime enforcement, is the goal.

Sanctum Runtime is an alternative open-core runtime project documented as MIT-licensed.

Verdict

Choose GLACIS if you need runtime decisions around AI actions plus signed, independently verifiable records while keeping payloads in your environment. Look elsewhere when automated evidence-pack generation, proof of correctness, complete event capture, or a broadly available managed rollout is more important than that combination.