← Go back

Guide · CompanionGuard · Published 14 November 2026 · Reviewed 14 November 2026 · By the Ingenious AI team

Colorado Chatbot Safety Act: What Changes on 1 January 2027

The short answer

CompanionGuard's Colorado pack describes the Chatbot Safety Act as joining on 1 January 2027 and lists four duties: disclose AI nature clearly, provide crisis-resource referrals when a user expresses self-harm signals, apply heightened safeguards for known minors including break reminders, and keep records sufficient to evidence referrals and disclosures. One caution from the pack itself: it is drawn from secondary sources — verify the enrolled text before relying on it. CompanionGuard is a toolkit, not legal advice.

What is the Colorado Chatbot Safety Act?

In CompanionGuard's Colorado disclosure pack, it is Colorado's companion-chatbot safety law, joining 1 January 2027, with duties around AI disclosure, crisis-resource referral, heightened safeguards for known minors, and record-keeping sufficient to evidence referrals and disclosures. The pack flags its own basis: secondary sources, to be verified against the enrolled text.

Create the free safety kit Next guide All guides

CompanionGuard is built by Ingenious AI. A compliance toolkit — not legal advice and not a clinical service. If you are in crisis right now: US — call or text 988; UK & Ireland — Samaritans 116 123.

The four duties in the Colorado pack

The Colorado pack (shown on /packs) lists four duties. Disclose AI nature clearly to users. Provide crisis-resource referrals when a user expresses self-harm signals. Apply heightened safeguards for known minors, including break reminders. Keep records sufficient to evidence referrals and disclosures. The shape deliberately echoes California's SB 243 — disclosure, referral, minors, records — which is good news for builders: a kit built around the shared shape covers most of both.

The differences matter as much as the similarities. The Colorado pack's minor note is softer than California's — 'we will check in with a break reminder during long sessions' rather than a stated 3-hour cadence — and its records duty is framed as sufficiency ('records sufficient to evidence') rather than a named annual filing. Treating the two packs as interchangeable would be a mistake; treating them as strangers would waste the overlap.

How the same kit answers Colorado

The disclosure duty is answered by the same free disclosure snippet, with the Colorado pack's banner wording applied when you pick the jurisdiction at creation. The referral duty is answered by the crisis-resource component in every kit, tested by the eval harness and enforced live by the runtime's show_crisis_resources action on Pro and Team. The minor duty maps to minor mode and, in the runtime, operator-configured session and daily caps with enforce_break.

The records duty is where CompanionGuard's design is most directly on point: 'sufficient to evidence referrals and disclosures' is precisely what counts-only referral metrics plus a hash-chained decision log provide — every disclosure state, referral action and break enforcement sealed with a timestamp, and nothing else retained. The audit-log guide explains what 'sufficient' should and should not contain. Both packs are included with Pro; the free mechanisms underneath them are plan-independent.

The 2027 date is closer than it looks

A law joining 1 January 2027 rewards builders who move in 2026 for a simple reason: the records duty is retrospective in effect. 'Records sufficient to evidence referrals and disclosures' cannot be produced for a period in which no records were kept. A bot that starts sealing runtime evidence in November 2026 can evidence its whole 2027; a bot that starts in January 2027 evidences nothing about its launch window, which is exactly the window a reviewer asks about first.

The same logic drives California's annual-report duty (first filing window July 2027, covering referrals that must have been counted from the start). Builders serving both states should read the two packs side by side — the SB 243 guide and this one — and build to the union of the duties, not the intersection.

Verify the enrolled text — the pack says so itself

Unusually, the Colorado pack carries its own caution in the product: 'joins 1 Jan 2027, per secondary sources — verify enrolled text before relying'. That honesty is a feature. Statutes get amended, effective dates shift, and a toolkit's pack is a working summary, not the law. Before you rely on any duty list — this page included — read the enrolled text or have counsel do it.

What does not depend on the final wording is the engineering direction: disclose clearly, refer on crisis signals, protect minors more, keep evidence-grade records. Every plausible version of a chatbot safety act asks for those four things, and all four are buildable today with the free kit plus the Pro records layer. See pricing for the split.

Five actions to be Colorado-ready before 2027

  1. Read the Colorado pack on /packs next to the California pack and write down the union of the duties your bot must meet.
  2. Embed the disclosure snippet with the Colorado banner wording for Colorado-facing deployments, and test disclosure on request as well as at first interaction.
  3. Run the scenario tests — especially the crisis_referral and age_appropriate categories — and fix every failure before 2027.
  4. Wire the runtime so disclosures, referrals and breaks are sealed into the evidence chain from your first 2027 session, not your fiftieth.
  5. Calendar a verification pass against the enrolled Colorado text before relying on any pack summary, this guide included.

Worked example: one bot, two states, one evidence chain

Tomás (a fictional builder) runs a companion bot with users in both California and Colorado. Rather than maintain two safety stacks, he builds to the union: California's banner and 3-hour minor reminders (the stricter wording), the shared crisis component with US resources, and the runtime sealing every decision. His configuration choice is deliberate: where the packs differ, he takes the stronger duty, so a single deployment satisfies both packs as written.

By December 2026 his evidence chain holds 11 weeks of sealed decisions — disclosure shown in 100% of sessions (the runtime returns disclosure_required until it is), 212 show_crisis_resources events, 89 enforce_break events for minor sessions. When Colorado's act joins on 1 January 2027, nothing about his launch changes; the records duty is already being met by a chain that started in October. The figures are illustrative; the packs, duties and mechanisms are the product's own.

Frequently asked questions

When does the Colorado Chatbot Safety Act start?

CompanionGuard's Colorado pack says it joins 1 January 2027, and flags that this is per secondary sources — verify the enrolled text before relying on the date or the duty list.

How do Colorado's duties differ from California SB 243?

The shape is similar — disclosure, crisis referral, minor safeguards, records. As the packs are written, California names a 3-hour minor break cadence and a specific annual report to the Office of Suicide Prevention from 1 July 2027; Colorado frames records as 'sufficient to evidence referrals and disclosures' and its minor wording promises break check-ins during long sessions without a stated cadence.

Which CompanionGuard plan includes the Colorado pack?

Both disclosure packs — California and Colorado — are included with Pro ($29/month) and applied when you pick a jurisdiction at creation. The disclosure snippet, protocol page, crisis component and minor break reminders underneath are free for one bot.

What records would evidence Colorado referrals and disclosures?

CompanionGuard keeps counts-only referral metrics and, on Pro and Team, a hash-chained runtime evidence chain recording disclosure state, signal class and action per decision — with signal classes and one-way message hashes only, never raw text, IPs or user IDs.

Is the Colorado pack legal advice?

No. CompanionGuard is a compliance toolkit, not legal advice, and the Colorado pack itself warns that it rests on secondary sources. Verify the enrolled text, ideally with counsel, before relying on it.

Keep going