NEW RESEARCH: Your Sandbox Is Made of Glass

Read

Trinitite

Continuous Audit

Audit once a year was built for systems that change slowly.

A model is retrained. A guardian is hot-swapped. A policy is finalized. By the time the auditor formalizes a workpaper, the system being audited is a different system. Trinitite turns audit into a stream — controls run on a schedule, produce signed evidence, and flow into your GRC tool, your auditor’s workpaper, and your underwriter’s portal from one chain of custody.

continuous_auditor · llm_rule

SIGNED

“No PHI in outbound LLM calls.”

compiled_to

llm_rule op

schedule

hourly

period_verdict

8,204 / 8,204 pass

dlir

e8d2…41aa · signed

cited_in

workpaper · chain_hash

“Can my underwriter see my posture today, without an NDA?” Yes.

The classical answer was “that requires a different engagement model.” Trinitite was built so the same primitives that drive a Big-4 engagement also drive a rolling assurance stream — internal audit, external audit, regulators, and underwriters consume it continuously.

The four primitives

Four services compose into one stream.

CCT

Continuous Controls Testing

Runs control tests on a schedule (SQL, Spark, dbt, API probe). Every run produces a signed row.

Internal audit, SOX / SOC 2 control owners

CCOE

Continuous Controls Operating Effectiveness

Each cycle Merkle-roots four signed inputs — CCT runs, chain integrity, TEE quote, guardian eval — into a KMS-signed workpaper auditors replay independently.

External audit, regulators, underwriters

Cont. Attestation

Continuous Attestation

A cursor-based, chunked verifier that signs anchor receipts as evidence accrues, anchored to RFC 3161 + Sigstore Rekor.

Auditors, regulators — no backdating possible

Auditor Agent

Continuous Auditor Agent

Express a rule in natural language; get a signed, deterministic verdict every period, DLIR-backed.

External & internal audit on novel risks

How it changes the engagement

Classical audit vs. the continuous stream.

Engagement once a year

Engagement augments the continuous stream

Sampling pulls from "the population we think exists"

Sampling pulls from a signed, attested population

Walkthrough = whiteboard + interviews

Walkthrough = signed CCOE assertions + cited DLIRs

"We relied on the SOC 2 report"

"We relied on continuous evidence anchored to Rekor at this chunk"

Annual SOC 2 / SOX cadence

Quarterly SSAE 21 / ISAE 3000 / SOC for AI from the same stream

In your language

Evidence in real time, in your terms.

Chief Audit Executive

Internal audit gets evidence in real time, not at year-end. Findings carry signed receipts at the moment of detection.

External audit (Big-4)

Continuous CCOE feeds the engagement workpaper; sampling pulls from a population already proven complete; replay is one click.

CISO

A live trust report you publish externally — your underwriter sees your posture today, without an NDA.

Reinsurer / cyber underwriter

Posture changes affect premium in real time; cession and warranty decisions cite signed continuous-audit evidence.

Continuous Audit is the always-on cut of the audit platform; the per-event stream is Continuous Assurance, and the kernel that makes the llm_rule honest is deterministic replay.

FAQ

Continuous audit, answered

What is continuous audit?

Continuous audit replaces the point-in-time engagement with a rolling assurance stream: controls run on a schedule, produce signed evidence, aggregate into rolling effectiveness assertions, get anchored to external clocks, and stream into your GRC tool, your auditor’s workpaper, and your underwriter’s portal — all from one chain of custody. Trinitite ships it as four primitives: CCT, CCOE, continuous attestation, and a continuous auditor agent.

How is continuous audit different from continuous assurance?

Continuous Assurance (Layer 2) is the per-event stream: every AI interaction signed and scored individually, rolled into one daily compliance number. Continuous Audit is the per-cycle control-evidence pipeline — workpapers, EQCR, signed attestation reports, OSCAL exports. They compose: a CCT cycle can include daily Continuous Assurance roll-ups as a controls-evidence input.

How can a natural-language audit rule be deterministic?

The platform compiles the auditor’s natural-language rule (e.g. "no PHI in outbound LLM calls") into a llm_rule op bound to a Trinitite guardian, runs it every period against that period’s calls, and signs the evaluation with a DLIR receipt. Because the kernel is deterministic, the same rule against the same evidence in the same period produces the same verdict — the auditor cites the DLIR and the signed verdict in the workpaper rather than trusting that the rule fired correctly.

What makes this honest where others can’t be?

Three things must compose: deterministic, signed inference for the llm_rule op (without it, AI-judged audit rules are an oxymoron); one chain of custody from telemetry to attestation; and external clock anchoring so silent backdating is impossible. See deterministic replay and the full audit platform.

Turn one control into a signed, rolling stream.

Stand up a CCT schedule on one control and watch the CCOE assertion, the anchor receipts, and the cited DLIRs accrue — workpaper-ready, underwriter-visible, no year-end scramble.

Trinitite

AI governance that catches mistakes, proves compliance, and shows the board what it saved—in dollars.

Trinitite is built by Fiscus Flows, Inc.

Products

Evals Continuous Evals Eval Harness ATLAS Red Team Guardian AI Reversible Masking MCP Governance CLI Firewall Latent Defense Skill Vault

Products

NHI Management Vector Integrity Trustworthy Knowledge Self-Improving Governance Shadow AI Inventory Risk Analytics Scenario Analytics Enterprise Reporting Compliance Connectors Pricing

Solutions

For Risk & Compliance For Model Risk (SR 11-7)For General Counsel For Engineering For Finance & Audit For CISOs For Insurers For Auditors For CPOs & Compliance

Resources

Audit Platform Continuous Audit Research Webinars & Videos Blog Podcasts AGRC Framework FAQ Architecture

Developers