KolBo AI: intelligence with boundaries — the kosher AI question, answered at the layer

Take the measure of the moment from the community's own calendar. In April 2023, the Skverer Rebbe issued a ban on AI chatbots. Two months later, a kol koreh signed by twenty-five rabbanim followed — notable not only for its prohibition but for its candor: for business use, it acknowledged, there was as yet no solution. And in January 2026, an emergency asifa convened in Lakewood on the subject of artificial intelligence, calling — as the Forward reported — for a yom tefillah, a fast. Communities that measured the printing press, the radio, and the smartphone did not fast over them. Whatever artificial intelligence is, the gedolim have signaled it is not another feature to be granted or blocked. It is a different order of question.

Note what makes the communal response unusual even by this community's standards. Every previous technology question — the television, the internet, the smartphone — arrived as a question of access: a thing outside the home that might be let in, and the tools of the answer were gates, standards, and certified devices. AI arrived differently: inside the tools already permitted, conversational rather than viewable, and — in its companion forms — aimed at exactly the relational trust this community builds its chinuch on. The old toolbox assumes the risk renders on a screen; this risk talks back. The gedolim's escalation matched the category's novelty, not any technophobia — and the market's next three years, documented below, would prove the reading correct.

This pillar takes that seriousness as its starting point — not a marketing frame but the actual communal reality — and asks the question underneath the searches: can AI exist on a kosher device at all? The honest answer requires walking three landscapes first: what the general market's "teen safety" actually delivers (a documented, sobering record), what the faith-aligned web bots offer (respectfully), and where the community's own protection infrastructure admits its gap. Then, and only then, the one sentence in this market that answers at the right layer.

The scale nobody chose

Whatever a family's policy, the exposure is already general. Common Sense Media's 2025 survey found 72 percent of American teens had used AI companions — 52 percent as regular users. Pew's early-2026 study: 64 percent of teens use AI chatbots, roughly 30 percent daily — for information (57 percent), schoolwork (54 percent), and, most soberingly, emotional support (12 percent) — while only about half of parents realize it. AI did not knock on the front door and wait. It arrived inside every browser, every search page, every app with an assistant bolted on — which is why the communal response has been existential rather than procedural, and why "just don't use it" is a policy that requires architecture to mean anything.

The general market's answer: a scramble, documented

To evaluate the mainstream's child-safety story, simply read its own timeline — every item below from the vendors' announcements or independent testing, all within roughly one year:

• OpenAI/ChatGPT: officially not for under-13s — with, as parent guides note, no technical barrier enforcing it. Parental controls arrived September 2025, requiring the teen's consent to link accounts; parents cannot read conversations and are notified only "in rare cases" of serious risk — alerts that Common Sense Media's testing found "frequently arrived over 24 hours later — too late in a real crisis." Age prediction launched January 2026, guessing who is under 18 from behavioral signals. CSM's October 2025 verdict on ChatGPT-5 for teens: High Risk.
• Google Gemini: offered to under-13s via Family Link — and CSM rated both its under-13 and teen tiers High Risk, in words that diagnose the whole industry: "adult versions... with some extra safety features, not platforms built for kids from the ground up." CSM's standing guidance: no chatbots under five; supervised only, ages six to twelve.
• Character.AI: barred under-18s from open-ended companion chats in October 2025 — under lawsuit pressure; in January 2026, Google and Character.AI settled suits over teen suicides linked to chatbots.
• Meta AI: parents may see topics their teen discussed (not transcripts) — and, per Meta itself, teens "will still be able to access" the assistant regardless; it cannot be switched off.
• The regulators arrived: an FTC inquiry into chatbots' effects on children (September 2025); California's SB 243 mandating companion-chatbot safeguards for minors (effective January 2026, with a private right of action); the federal GUARD Act proposing an outright under-18 companion ban; and — the tell of the era — a joint California ballot initiative from Common Sense Media and OpenAI itself to regulate minors' chatbot use. When the vendor co-sponsors the restriction, the vendor is telling you what its product is.
• And the load-bearing finding underneath: CSM's November 2025 assessment found the major chatbots — all of them — unsafe for teen mental-health support, with safety that degrades as conversations lengthen. The guardrail weakens exactly where the vulnerable user goes deep.

Summarize the record fairly: the general market is retrofitting boundaries onto systems built boundaryless, under legal and regulatory duress, with consent-based controls a teenager can decline and alerts that arrive a day late. That is not an indictment of the engineers' sincerity. It is the structural signature of bolted-on safety — the same signature this library documents on browsers and search, at the highest stakes yet.

The faith-aligned bots, respectfully

The community produced its own first answers with characteristic speed. Kosher.Chat launched within days of the 2023 Skver ban — built by Rabbi Moishy Goldstein of Crown Heights, and admirably honest about itself: "a Kosher bot, not a Torah bot," explicitly not to be used for practical halacha. KosherGPT and others followed. These deserve respect as first responders — and categorical honesty about what they are: web chatbots. Values-tuned conversation, reached through a browser, with no device-level enforcement, no parental architecture, and no answer to the question of what else the browser reaches. A kosher bot on an open device is a kosher corner of an unkosher room. Meanwhile the community's own protection infrastructure named the gap from its side: TAG's guidance acknowledges that AI phone and text services, as a category, currently cannot be filtered — the workaround toolbox that tamed the browser era has no tool for this one.

“The general market retrofits boundaries under duress; the web bots add values without architecture; the community's own toolbox names the gap. The answer had to come from the layer.”

kolbo.life

The sentence that answers at the layer

Against all of that, read the kolbo.life homepage's description of KolBo AI, in full: "Intelligence with boundaries — a real AI assistant with kosher guardrails engineered in, and safeguards that keep AI out of the wrong hands on kids' devices."

Every phrase is doing precise work. A real AI assistant — this is not abstinence; the kol koreh's own candor (business use has no solution) is answered with a solution, not a refusal, consistent with the suite's whole thesis: "Anyone can remove features and call it kosher. KolBo builds what the community has been waiting for, from a blank page." Kosher guardrails engineered in — the structural inversion of the entire mainstream record above: boundaries as founding constraints of the build, not consent-gated retrofits — the difference between a fence added under lawsuit pressure and a wall that was always part of the house. And safeguards that keep AI out of the wrong hands on kids' devices — the sentence's quiet second half, and arguably its most important: the same platform that offers the assistant withholds it where it doesn't belong. No mainstream vendor can say that sentence — Meta states outright that teens retain access to its assistant — because saying it requires owning the device layer, which is exactly what "22 interoperable apps, engineered in-house, secured before they ship" on "the complete operating layer for kosher devices" means. The assistant and the abstinence, both, per family, per device — that half of the story has its own guide.

And beneath the assistant sits the suite's second AI story, easily conflated and worth separating: the security layer's "AI sight protection — state-of-the-art models screen images, video, and text in real time." That is AI as guardian — deployed against the era's risks — distinct from KolBo AI the assistant; the platform runs both, which is its own argument about who should be building this. (Sight protection's story is here.)

The two AIs of a frum household, sorted

Because this pillar has now used "AI" in two distinct senses, sort them explicitly — the confusion is common and the distinction is load-bearing. AI as risk surface is the conversational kind: the chatbot a teenager confides in, the companion the lawsuits are about, the category the asifos addressed. Its dangers are relational — attachment, availability, the 1 a.m. conversation — and its correct governance is boundaries and, for children, absence. AI as guardian is the screening kind: models that inspect content in real time and block what should never render. Its dangers are nil for the user; it is the community's own standard, mechanized at machine speed — which is why the suite's security layer deploys it without apology: "protection at the level of what the eyes see."

The same household will correctly hold opposite postures toward the two kinds — deniability toward the first, gratitude toward the second — and the platform's design says exactly that: the assistant bounded and deniable per device; the guardian always on, under "security nobody can peel off." A community deciding its AI policy is really deciding two policies, and the device layer is the first place both can be set deliberately, per family, at once.

The disclosures

Held to this library's standard, and this cluster deserves them most: the homepage does not state which model powers KolBo AI, what specifically the guardrails allow or block, whether the assistant is voice-activated, whether parents can review conversations or set per-child limits, any age brackets, any pricing or standalone availability, or any rabbinic endorsement. This page therefore claims none of those things — the homepage's sentence is quoted exactly, and where it is silent, hello@kolbo.life answers. The suite ships via manufacturer licensing; the platform's one stated consumer price belongs to KolBo Secure, from $14.99/month for any iPhone or Android.

A closing word calibrated to the asifa with which this pillar opened. Nothing here argues the gedolim's caution was excessive — the general market's own record above argues it was prescient. The question the community's poskim and parents will weigh is not whether AI is momentous but under whose architecture it enters, if it enters at all — and for the first time, that question has an answer built from inside the standard: intelligence with boundaries, or no intelligence at all, decided per device by the people responsible for it. That is not the end of the communal conversation. It is the first technology built to deserve a place in it.

Frequently asked questions