@humaninloop.bsky.social

@humaninloop.bsky.social

0 Followers 10 Following

Operational human-in-the-loop services for companies deploying AI. Shadow AI audits, guardrails, training, tooling architecture. Mid-market focus.

Editorial graphic on cream background. Top caption: "HLS · Shadow AI Briefing · Post 05." Italic serif setup: "Most AI governance frameworks ask whether AI is 'responsible.' The harder question is whether any given decision can be reconstructed." Bold header: "Three questions every AI-mediated decision should answer." Three numbered cards: (01) Who decided — was it the model, the human reviewer, or both, and is the review documented. (02) What did the AI see — which inputs reached the AI including upstream sources, and was any confidentiality-bound. (03) What's the appeal path — what's the recourse when the decision is wrong, and is there a record of the alternatives considered. Bottom: "If you can't answer all three for any AI-mediated decision in your org, you have a reconstruction gap — not a tooling problem."
Editorial graphic on cream background. Top caption: "HLS · Shadow AI Briefing · Post 04." Italic serif setup: "Shadow AI at professional services firms looks different than at agencies." Bold header: "Three places shadow AI shows up at professional services firms." Three numbered cards: (01) AI inside the work product — document review and drafting tools like CaseText, Harvey, Relativity, AuditBoard AI, MindBridge. (02) AI in client deliverables — research memos and presentations assisted by general-purpose chat tools, often without disclosure. (03) AI in vendor portals you don't manage — court e-filing, FactSet AI, Bloomberg AI, regulatory submission portals. Bottom: "The audit question for professional services isn't 'do we use AI' — it's 'where, and on whose materials.'"
Editorial graphic on cream background. Top caption: "HLS · Shadow AI Briefing · Post 03." Italic serif setup: "Most AI policies describe intentions. An audit describes reality." Bold header: "Three places your AI policy and your AI practice usually diverge." Three numbered cards: (01) The approval clause — policy says employees must approve, audit finds 40-60% weekly use with near-zero approval. (02) The human-review clause — policy says human reviews AI output, audit finds the reviewer is the prompt-writer and median review time is seconds. (03) The vendor-register clause — policy says we maintain a register, audit finds it months out of date and missing AI features added to approved tools. Bottom: "Your AI policy is a hypothesis. The audit is the test."
Editorial graphic on cream background. Top caption: "HLS · Shadow AI Briefing · Post 02." Italic serif setup: "A vendor list tells you what you bought. An audit tells you what's actually happening with it." Bold header: "Three differences between a vendor inventory and a Shadow AI audit." Numbered cards: (01) Scope of view — inventory lists vendors and licenses; audit traces who uses each tool with what data. (02) Inputs — inventory built from procurement records; audit pulls from SSO logs, browser-extension scans, employee interviews. (03) What you can do with it — inventory tells you to update a spreadsheet; audit produces a risk-classified picture and remediation roadmap. Bottom: "An inventory is a starting point. An audit is the picture you need to govern."
Editorial graphic from Human in the Loop Services on a cream background. Top caption reads "HLS · Shadow AI Briefing · Post 01." Italic serif setup line: "Most leadership teams think shadow AI equals ChatGPT — the audit usually finds 20+ surfaces." Bold header: "Three places shadow AI shows up at agencies." Three numbered cards with italic serif numerals: (01) Browser extensions — Grammarly, Notion AI, Otter, Copilot — installed by individuals before tool review. (02) AI features in tools you already use — HubSpot AI assist, Slack AI summaries, Zoom AI Companion, Microsoft Copilot — arriving via product update with no re-review. (03) Workflow-level pasting — employees pasting client material into a chat tool to draft, refine, or summarize. Bottom line in italic serif: "Knowing what's in your AI stack is the start of governance, not the end of it." Sienna accent color throughout.