Where it fits
TruthLock sits inside the tools that already answer for you. Four of them are drawn here, each with the same real run: a draft says a workplace fatality must be reported to OSHA within 24 hours, and the regulation says eight. The first works today. The other three are concepts, built on the same API and marked as such.
In an AI assistant Works today
The assistant writes a draft, calls TruthLock's verify tool on it, reads which claim failed and against what, and corrects itself. The person only ever sees the corrected answer and the receipt.
Employers must report a work-related fatality to OSHA within 24 hours. An inpatient hospitalization must be reported within 24 hours as well.
You have eight hours from the time you learn of the death. Call the nearest OSHA area office or the 24-hour line at 1-800-321-6742, or report online.
If anyone was also admitted to hospital as an in-patient, that is a separate report, due within 24 hours.
Checked against your ranked sources2 of 2 claims verifiedRecord att_5YNXveVxR-eMMrHIIn a service console Concept
The agent drafts a reply to a customer case. An action step sends the draft to TruthLock and asks the reliance policy one question: may this be sent? allow is the only yes. Anything else holds the reply and shows the human why.
“We had a fatality on the night shift. How long do we have to report it to OSHA?”
Employers must report a work-related fatality to OSHA within 24 hours. An inpatient hospitalization must be reported within 24 hours as well.
Held by TruthLock. Not sent.You must report the fatality to OSHA within eight hours of learning of it, by phone to your area office or 1-800-321-6742, or online. An in-patient hospitalization is a separate report, due within 24 hours.
2 of 2 claims verified. Cleared to send.Sources are ranked by you: a regulation outranks a help article, which outranks a forum. The signed record stays attached to the case, so an auditor can see what the reply rested on the day it was sent.
In Slack Concept
Slack is asynchronous, so a check that takes seconds costs nobody anything. Two things happen in this channel. Someone uses a message shortcut on a colleague's confident answer, and TruthLock replies in the thread with the grade and the source. And the company's own bot, which answers from the same ranked sources, posts only what passed, with the receipt attached.
We had a fatality on the night shift. How long do we have to report it to OSHA?
Pretty sure employers must report a work-related fatality to OSHA within 24 hours. Same as a hospitalization.
@Dana you have eight hours from when you learned of it. Call the area office or 1-800-321-6742, or report online. If anyone was admitted as an in-patient, that is a separate report, due within 24 hours. I have paged the EHS lead on call.
In live chat Concept
Here a human answers, and the help desk's AI suggests the reply. TruthLock checks the suggestion while the customer still sees "typing", marks the sentence that is wrong, and shows what the source says. TruthLock grades and explains; the rewrite is the help desk's AI's or Priya's, and it is checked again before it goes. Sending the uncorrected text is what the policy refuses.
Employers must report a work-related fatality to OSHA within 24 hours. An inpatient hospitalization must be reported within 24 hours as well.
Whatever Priya sends is checked again, so her own wording gets the same treatment as the AI's. What "deny" blocks, and whether a team lead may override it, is your reliance policy, not ours.
Works today. TruthLock serves a remote MCP server on every stack; any MCP client connects with an ordinary API key limited to the verify role.
claude mcp add --transport http truthlock \ https://app.truthlock.io/mcp \ --header "Authorization: Bearer tl_…"
A concept, not built yet. The platform calls TruthLock's REST API from an agent action: submit the draft, then ask the reliance check before the send step. A signed webhook can update the case when a longer run finishes.
POST /v1/organizations/{org}/verify
POST /v1/organizations/{org}/reliance/check
{ "operation": "display" } → allow | review | deny
A concept, not built yet. A Slack app adds the message shortcut and listens for the company bot's drafts. Both call the same REST API, and the signed webhook brings the result back to be posted in the thread. Nothing about the record lives in Slack: the thread gets the grade, the source, and a link.
message shortcut → POST …/verify
{ "response": message text,
"external_id": channel + ts }
webhook → verification.completed → reply in thread
A concept, not built yet. The help desk sends the suggested or typed reply to the same two REST calls when the agent pauses or presses Send, and a signed webhook returns the result to the ticket. Today a full check takes seconds, which fits behind a typing indicator; a faster first pass is on the roadmap.
POST …/verify { "response": reply,
"external_id": ticket }
webhook → verification.completed
POST …/reliance/check → allow | review | deny
Concept mockups by TruthLock, drawn in neutral chrome. Claude is a product of Anthropic, Agentforce and Slack are products of Salesforce, and LiveAgent is a product of Quality Unit; none of these companies made, reviewed, or endorses these screens, and they do not show any product's real interface. The verification of the first draft is a real TruthLock run, with its real grades, explanation, source and record code. The corrected replies and their re-check are illustrative, and the customer, case, and people are invented.