# Take Interest Inc. > Take Interest Inc. helps you finish what you started with AI. Founded 2025, a US C-Corp at takeinterest.ai. The team that builds an AI system should not be the one who says it is safe, so the live service is one published rung: a one-week security and governance review of what a company AI system can already do on its own, meaning what it can reach, what it can change, and what it can do with nobody watching. The review is one fixed scope at one fixed fee, agreed on a free thirty-minute call before any work starts, and no fee figure is published on the site or in this file. Fixing what the review finds is quoted per client afterwards, and keeping the gates running is a separate subscription. It ends in two things: a written capability map ranking what to fix first, with anything that could not be checked marked not measured rather than passed, and a live walkthrough where the owner watches their own AI act. Fixing what the review finds is quoted per client after the review rather than published as a band. A subscription that runs every release through the same gates is built and priced. Five checks run over the system (static analysis, dependency vulnerabilities against OSV, injection scanning of AI instruction files, secret scanning, and an audit of how AI answer tools describe the business) and about eight founder hours go into the read and the walkthrough. The same gates clear every commit, release and outbound send at this company. Both small businesses and large enterprises are in scope. $2M of professional liability sits behind the opinion, on top of that. We also build products: GuardClaw is runtime security for tool-using agents, available now in public beta. Data stance: you decide what is collected, you control it while it is here, and you can export and take it with you. Every person, free or paid, can turn off data collection, and a paid collects-nothing tier stays available as a product choice. Every figure we publish traces to a recorded measurement in our own repository; we publish no client outcome averages. ## Reading the Blog Every blog post is served as semantic HTML with JSON-LD `` payload for crawlers. The HTML view is the only public projection — the prior per-post `.md` and `.json` endpoints were removed 2026-05-24 to protect the editorial work (text + audio + video + mind map + infographic are a curated combination, not a downloadable corpus). Citing crawlers can parse the HTML directly or use the embedded JSON-LD. Canonical URL shape: `https://takeinterest.ai/blog/` (e.g. `/blog/30-day-agent-security-checklist`). ## Products - [GuardClaw, Runtime Security for Tool-Using Agents](https://takeinterest.ai/guardclaw): 7 layers of deterministic runtime security. Enforces policy rules rather than probabilistic inference. Security decisions run locally and your code is never uploaded. Anonymous telemetry and a pattern update check run by default and can be turned off. Pattern updates and optional receipt sync do reach our servers, and you can turn both off. 1,600+ detection patterns. Available now (public beta, free). MCP-native, works as a security gateway for any MCP client. - [GuardClaw Documentation](https://takeinterest.ai/guardclaw/docs): Quickstart, configuration, policy authoring, threat model, architecture reference. - [Take Interest, agent-ready context for people and their agents](https://takeinterest.ai/thinkingstrategy): Captures the working knowledge people build, keeps it private to each person, checks it before anything acts, and teaches the person to use it well. In testing. - [Askwell](https://takeinterest.ai/askwell): A local-first iPhone app for asking sharper questions. It scores a question on device, names missing context, suggests a clearer version, and saves a proof note. Pre-launch; not on the App Store yet. - [Felt Weather](https://takeinterest.ai/felt-weather): An iPhone-first activity and event planner that shows how the day will feel and what to move. Weather is an input to the plan, not the category. Free for the first 14 days on the App Store for iPhone, iOS 17 and later, then Felt Pro at $9.99/month. - [Pantry Kitchen Journal](https://takeinterest.ai/pantry): Take Interest applied to the kitchen. Scan a receipt or barcode, see what is on the shelf, and cook from what is already there. Available on the App Store for iPhone. ## Key Pages - [Services](https://takeinterest.ai/services): The live service and what it costs. AI for small business: use-case identification, workflow setup, AI training for employees, and onboarding systems. Starts with a free 30-minute diagnostic. One-week security and governance review, one fixed scope at one fixed fee given on that call. The build that fixes what the review finds is quoted per client after the review. - [AI adoption for small business](https://takeinterest.ai/ai-adoption-for-small-business): Who the service is for and how the engagement runs, for a small or legacy business that has not adopted AI yet. - [Security Posture](https://takeinterest.ai/security): How we build and operate. Zero trust architecture, data handling, compliance. - [Verify a signed report](https://takeinterest.ai/verify): Our public signing key (ECDSA P-256 over SHA-256, key id 9f7d56d92d05f139b569f8e6be3eaaf7, published at /attestation/) and the two openssl commands that confirm a delivered report came from Take Interest and has not been altered. Checkable with no account and no Take Interest software. States plainly what a valid signature does not prove: not a grade on the code, not proof that everything was examined, and not an independently notarised date. - [About](https://takeinterest.ai/about): Company, mission, team. - [Blog](https://takeinterest.ai/blog): The Builder's Guide to Agent Security plus the small business AI playbook. 98 posts as of 2026-09-11, covering agent threats, zero trust, permission scoping, AI adoption, sales practice, and founder operations. Full index: https://takeinterest.ai/llms-full.txt - [Contact](https://takeinterest.ai/contact): General inquiries, security reports, partnerships. - Account and progress routes are member utilities, not discovery pages. They are intentionally outside the sitemap and carry `noindex`. ## GuardClaw MCP Integration - MCP Server Name: `guardclaw` - Transport: stdio (local), streamable-http (cloud) - Tools: `shell` (protected command execution), `filesystem` (protected file operations), `http` (protected HTTP requests) - Security layers: Threat Intelligence, Input Validation, Policy Enforcement, Capability Tokens, Sandboxed Execution, Human-in-the-Loop, Receipt Chain - Server Card: [.well-known/mcp.json](https://takeinterest.ai/.well-known/mcp.json) ## Blog — Agent Security - [Your AI Agent Has No Seatbelt](https://takeinterest.ai/blog/your-ai-agent-has-no-seatbelt): AI agents ship into production faster than safety standards can keep up. Runtime security controls are overdue. - [The Identity Problem](https://takeinterest.ai/blog/the-identity-problem): Non-human identities outnumber human users in enterprise environments, yet most orgs manage agent credentials like shared passwords. - [Everyone's Worried About Prompt Injection. That's the Easy Problem.](https://takeinterest.ai/blog/prompt-injection-is-the-easy-problem): Six other agent attack vectors cause more damage. Mapping your full attack surface takes 30 minutes. - [Zero Trust Was Built for Humans](https://takeinterest.ai/blog/zero-trust-was-built-for-humans): Zero trust principles hold for AI agents, but the implementation needs a complete rethink. - [Build Like You'll Get It Wrong](https://takeinterest.ai/blog/build-like-youll-get-it-wrong): Resilience beats perfection. Design recovery into every system. - [Security Is a Primitive, Not a Feature](https://takeinterest.ai/blog/security-is-a-primitive): Security is load-bearing architecture. Three primitives every agent system needs before first deploy. - [Seven Layers of Defense](https://takeinterest.ai/blog/7-layers-of-defense): What defense in depth actually looks like for AI agents, and why edge checks alone leave the middle unguarded. - [Why We Don't Use AI to Make Security Decisions](https://takeinterest.ai/blog/why-we-dont-use-ai-for-security-decisions): Probabilistic models for detection and triage, deterministic rules for enforcement. No LLM in the deny/allow path. - [The 30-Day Agent Security Checklist](https://takeinterest.ai/blog/30-day-agent-security-checklist): Four weeks from "we should secure our agents" to a tested, documented security posture. - [Operator Readiness Assessment](https://takeinterest.ai/blog/operator-readiness-for-ai-workflows): A 15-minute self-assessment across five dimensions. - [The Builder's Responsibility](https://takeinterest.ai/blog/the-builders-responsibility): We're in a cathedral-building moment for AI. Today's safety decisions shape decades of autonomous systems. - [Why We Built GuardClaw](https://takeinterest.ai/blog/why-we-built-guardclaw): AI agents moved from demos to operators. The threat model changed faster than defenses. - [820 Malicious Agent Skills and Nobody Noticed](https://takeinterest.ai/blog/820-malicious-agent-skills-nobody-noticed): Koi Security found 820+ malicious skills on ClawHub. Agent marketplaces are the new attack vector. - [88% of AI Agents Shipped Without Security Sign-Off](https://takeinterest.ai/blog/88-percent-agents-shipped-without-security): Gravitee 2026 data: only 14% got full security approval before deploying agents. - [Agent Supply Chain Security in 5 Steps](https://takeinterest.ai/blog/agent-supply-chain-security-five-steps): From skill vetting to dependency pinning to runtime monitoring. - [4.5x More Incidents Start with One Setting](https://takeinterest.ai/blog/agents-with-admin-access-4-5x-more-incidents): Teleport's 2026 research: over-privileged agents experience 4.5x more security incidents. - [Audit Your Agent's Trust Boundaries This Week](https://takeinterest.ai/blog/audit-agent-trust-boundaries-this-week): Map and test every trust assumption your AI agents make. - [Designed to Work Together From Day One](https://takeinterest.ai/blog/designed-to-work-together-from-day-one): Why disconnected AI tools create hidden risk. - [70% of Enterprises Can't See Their Own Agents](https://takeinterest.ai/blog/enterprises-cant-see-their-own-agents): Most enterprises can't tell you how many agents they run, what they access, or who owns them. - [5 Things Due Before August 2 — EU AI Act Checklist](https://takeinterest.ai/blog/eu-ai-act-checklist-before-august): The EU AI Act high-risk deadline actions you can start this week. - [One Firebase Misconfig Leaked 300M Chat Messages](https://takeinterest.ai/blog/firebase-misconfig-leaked-300m-messages): An AI chat app left a Firebase database open. 300 million messages exposed. - [The Friday Agent Permission Audit](https://takeinterest.ai/blog/friday-agent-permission-audit-checklist): 90-minute permission audit. Nine checks, one agent at a time. - [How Fast Can an Attacker Hijack Your Agent?](https://takeinterest.ai/blog/how-fast-can-attacker-hijack-agent): CrowdStrike says attack timelines are under 72 minutes. - [Innovation and Security Are the Same Product Decision](https://takeinterest.ai/blog/innovation-and-security): Treating security as architecture accelerates shipping. - [Least Privilege Wasn't Built for Agents](https://takeinterest.ai/blog/least-privilege-wasnt-built-for-agents): The principle holds. The implementation breaks when the user makes 10,000 decisions per hour. - [Microsoft Found a New Way to Poison AI Recommendations](https://takeinterest.ai/blog/microsoft-recommendation-poisoning-attack): Summarize buttons can be weaponized. Recommendation poisoning is the supply chain attack nobody planned for. - [NIST Wants to Know How You Secure Your Agents](https://takeinterest.ai/blog/nist-ai-agent-security-rfi-breakdown): NIST AI Agent Standards RFI breakdown. What it asked, what it signals. - [NIST Wants Agents Governed Like Employees](https://takeinterest.ai/blog/nist-wants-agents-governed-like-employees): NIST signals agents need identity, accountability, and lifecycle management. - [One Localhost Assumption Gave Hackers Full Control](https://takeinterest.ai/blog/one-localhost-assumption-gave-hackers-control): The ClawJacked vulnerability — a single implicit trust assumption let any website take over a machine. - [Prompt Injection Just Got Classified as Malware](https://takeinterest.ai/blog/prompt-injection-classified-as-malware): Researchers want prompt injection reclassified. A $40K bounty is testing why. - [Three Layers of Agent Permission Scoping](https://takeinterest.ai/blog/three-layers-agent-permission-scoping): Identity (who), scope (what), context (when). How to build all three. - [We Trust Systems We Can't Inspect Every Day](https://takeinterest.ai/blog/we-trust-systems-we-cant-inspect): From plumbing to power grids to AI agents — invisible infrastructure trust. - [What Happens When Agents Outnumber People?](https://takeinterest.ai/blog/what-happens-when-agents-outnumber-people): Machine identities outnumber humans 25-50x. Governance frameworks built for human-majority orgs are expiring. - [What New Hires and AI Agents Have in Common](https://takeinterest.ai/blog/what-new-hires-and-ai-agents-have-in-common): Your company onboards people. It probably doesn't onboard agents. Same trust-building patterns apply. - [Why Your Agent Has More Access Than You](https://takeinterest.ai/blog/why-your-agent-has-more-access-than-you): 70% of security leaders say agents have more access than humans in the same role. ## Blog — Agents as Customers - [Your Next Customer Is an Agent](https://takeinterest.ai/blog/your-next-customer-is-an-agent): Agents are becoming buyers, not just readers. They carry an identity, a scoped wallet, and an inbox. What it takes to be the business an agent can find, trust, and pay. Agent-ready is not scrape-ready: an agent acts for a specific person, inside that person's consent. ## Reading Paths ### Getting ready for agent customers Your Next Customer Is an Agent → llms.txt (this file) → .well-known/mcp.json → Security Posture → The Identity Problem ### Exploring the Take Interest product family Take Interest → Askwell → Felt Weather → Pantry → Security Posture → Contact ### Securing AI agents (start here) Your AI Agent Has No Seatbelt → Prompt Injection Is the Easy Problem → Seven Layers of Defense → Why We Don't Use AI for Security Decisions → GuardClaw → Docs ### Understanding the threat model The Identity Problem → 820 Malicious Agent Skills → How Fast Can an Attacker Hijack Your Agent → One Localhost Assumption → Zero Trust Was Built for Humans ### Operationalizing agent security 30-Day Agent Security Checklist → Friday Agent Permission Audit → Audit Your Agent's Trust Boundaries → Operator Readiness Assessment → Three Layers of Permission Scoping ### Compliance and governance EU AI Act Checklist → NIST Wants Agents Governed Like Employees → NIST AI Agent Standards RFI → What Happens When Agents Outnumber People ### How GuardClaw works Why We Built GuardClaw → Security Is a Primitive → Seven Layers of Defense → Why We Don't Use AI for Security Decisions → What We Got Wrong → GuardClaw Docs ## Core Principles - Security is a primitive, not a feature. It goes in before the first line of code. - Deterministic enforcement. No language model sits in the security decision path. - Consent, control, and portability. GuardClaw is zero-data-out and runs in your infrastructure. Take Interest is cloud-first on Google Cloud at launch, with a progressive path to on-device by V3. You decide what is collected, you control it while it is here, you can turn collection off at any time, and you can export and take your data with you whenever you leave. - Process over outcomes. Fix the process, not the people. - Show the work, including the wrong turns. ## Feeds - [RSS](https://takeinterest.ai/rss.xml) - [Sitemap](https://takeinterest.ai/sitemap-index.xml) - [Discord](https://discord.gg/XkfNcJ9JY) - [LinkedIn](https://www.linkedin.com/company/take-interest-inc/) - [Loop](https://app.ourloop.life/loop/take-interest) ## Legal - [Privacy Policy](https://takeinterest.ai/legal/privacy) - [Terms of Service](https://takeinterest.ai/legal/terms) - [GuardClaw Privacy](https://takeinterest.ai/guardclaw/privacy) - [GuardClaw Terms](https://takeinterest.ai/guardclaw/terms) - [Pantry Privacy](https://takeinterest.ai/pantry/privacy) - [Felt Weather Privacy](https://takeinterest.ai/felt-weather/privacy) - [Askwell Privacy](https://takeinterest.ai/askwell/privacy) ## Contact - General: hello@takeinterest.ai - Security: security@takeinterest.ai - Privacy: privacy@takeinterest.ai