Agents already hold the repo, the database, the money. Permission never shipped with them. Wrapbox answers every agent action before it runs — and proves every answer after.
Agents got the access.
Nobody signs off on the act.
The same access as an employee — code, customer data, money. But when an agent acts:
No approval. No size limit. No proof.
One API call by a coding agent. A startup’s production database and every backup — gone. Nobody asked it to delete anything.
of breached orgs with an AI incident had no AI access controls
IBM · Cost of a Data Breach 2025 ↗of enterprises run AI agents nobody registered
Cloud Security Alliance · Apr 2026 ↗of successful agent attacks through 2029 will exploit access-control gaps
Gartner · Aug 2026 ↗They check the tool’s name.
Wrapbox reads the whole act.
A permit layer beneath your agents — nothing to hook, nothing an agent can switch off. Each consequential act is read on five facts, together — and answered one of four ways.
security products benchmarked against Wrapbox
already claim a check before the action runs
read all five facts — or send the sign-off to a different person than the requester
2021 — nothing to guard. 2029 — a checkbox.
2026 — the window.
Agents could only suggest text. No MCP, no acting agents — and no “securing‑AI” budget line existed.
Agents act with real credentials on laptops and in production. Public incidents give CISOs a reason to act. Incumbent controls shipped months ago — key parts still in preview.
“Good‑enough” agent controls bundled into platforms; Gartner expects consolidation. A 2029 startup competes with a free checkbox.
Shipped into the window: Microsoft Agent 365 · May 2026 — Delinea runtime authorization · Jul 2026 — CrowdStrike Falcon Guardian · Sep 2026.
monthly MCP SDK downloads — one year after launch; agents now act on real systems
MCP · Anthropic · Dec 2025 ↗of organizations say their AI agents already went beyond intended permissions
Cloud Security Alliance · Apr 2026 · Zenity‑funded ↗securing‑AI spend next year, to ≈$4.8B in 2027 — a budget line that did not exist
Gartner forecast · Aug 2026 ↗Every large company will need a safety check
before its agents act.
Per year · bottom‑up · not to scale
Every company with 500+ staff in the US, EU and UK — 21,041 US, ≈23,000 EU, ≈3,500 UK, from official statistics — at Wrapbox’s own ACVs: $50k, $162k or $558k by company size, ≈$150k blended.
24,000 companies with 1,000+ staff × McKinsey’s 23% that already run AI agents at scale — not the 62% that also counts “experimenting” — is 5,500 companies, at a $245k blended ACV.
Year 5: 300 customers × $150k = $45M ARR — 5.5% of SAM companies, a small share of a category still picking vendors. Wiz, the outlier, passed $100M ARR within 18 months of its first $1M.
Two methods agree within 2×. Per seat: 14.6M professional developers × 62% on AI coding tools × $15–25 a seat a month = $1.6–2.7B — between the SAM and the TAM. SlashData · JetBrains 2025 ↗
$150k is what security teams already pay per tool. Vendr medians: Okta ≈$45k, CrowdStrike ≈$56k, Zscaler ≈$80k, Netskope ≈$97k. Vendr ↗
Gartner’s top‑down forecast: ≈$4.8B on securing AI in 2027, up 69% — kept as the cross‑check, not the TAM. Gartner · Aug 2026 ↗
US Census SUSB 2022 via SBA Advocacy (Feb 2026) · Eurostat SBS (Dec 2025) · UK DBT Business Population Estimates 2025 · McKinsey State of AI (Nov 2025) · SlashData (May 2025) · JetBrains (2025) · Vendr · Gartner (26 Aug 2026) · Wrapbox pricing.
148 companies checked. 89 can stop an agent before it acts. None ask where the data goes, or how much could break.
Click any name on the map. 01–05 are the five strongest competitors.
Wrapbox Global Competitive Benchmark, 26 Sep 2026 — 7 research streams, 560 source URLs (vendor docs, release notes, repos, press releases, filings, opened directly), 193 verification-log items. Positions are the researchers’ 1–5 scores; 89 of 148 rows coded pre-execution; 21 unscored rows not plotted.
Watch it run.
A full walkthrough of the real Wrapbox prototype, screen by screen.
Pay per person and per agent.
Every check included.
- One laptop
- Free forever
- 25–250 people
- One place checked
- Every place checked
- Sign‑off · full record
- $50k a year minimum
Planned list prices · to be tested with design partners
The unusual part. Microsoft charges $15 a person and includes every agent — its revenue stays flat as agents multiply. We also charge per agent, so we grow as they do. Agents with no human owner fit no seat; no one publicly prices blocking them yet.
Per‑check pricing is a race to the bottom — AWS charges $0.025 per 1,000 — and gives buyers bills they can’t forecast. We include 100,000 checks per person or agent a month. The buyer gets a flat bill; we are paid for safety, not compute.
$25k for 60 days, 100% credited to year one — the buyer risks little. The yearly order form is signed before the pilot starts, so a good pilot becomes a contract without a second sale.
A 1,000‑staff company: 300 people × $25 × 12 = $90k, plus 200 agents × $12 × 12 = $28.8k. $119k a year at list, about $95–100k after the usual discount.
$15 a person is what Microsoft charges for Agent 365. Per‑person prices cluster at $8–15 a month (Obsidian, Harmonic, WitnessAI). Enterprise at $25 has to earn its premium by blocking, not only watching. AWS Marketplace ↗
Specialist AI‑security platforms list at $100–250k a year (Straiker, Pillar, Island). Our $50k minimum sits below what the median company pays CrowdStrike or Zscaler — easy to approve.
Wrapbox pricing & business‑model research, 26 Sep 2026 — 45 competitors checked · AWS Marketplace listings (WitnessAI, Harmonic, Obsidian, Straiker, Pillar, Island) · Microsoft Agent 365 (GA 1 May 2026) · AWS Bedrock AgentCore pricing (25 Sep 2026) · CostBench / Vendr contract medians. Prices are planned list prices, to be tested with design partners.
Earn trust with research.
Win with design partners. Scale with partners.
Name the attack, give away the finder. Invariant Labs, about 10 people, named “tool poisoning” in April 2025 and shipped a free scanner — Snyk bought it that June. We start the same way, before we raise.
3–5 regulated companies, 500+ staff, already on Claude Code, Cursor or Copilot. Free early access and a weekly founder call, in return for feedback. Keycard launched with “a handful”; 7AI’s design partner later became its reseller.
Design partners become $25k, 60‑day paid pilots, then yearly contracts. Security is bought through partners — Canalys expects over 90% of 2025 cybersecurity spend to involve one — so AWS / Azure Marketplace and resellers come last.
Johann Rehberger · Independent researcher, Embrace The Red — his “Month of AI Bugs” disclosed two dozen+ flaws in Claude Code, Copilot, Cursor and others (Aug 2025) — the exact actions we stop.
Simon Willison · Independent developer, coined “prompt injection” — argues coding agents should run in a sandbox by default — and amplifies good agent‑security research.
Clint Gibler · tl;dr sec newsletter, 50,000+ subscribers — covers agent security weekly and picks up useful open‑source tools.
Ashish Rajan · Host, Cloud Security & AI Security Podcast — 2026 episodes on over‑privileged agents and agent authorization.
Caleb Sima · Co‑host, AI Security Podcast — reaches the security leaders who will buy this.
David Spark · Producer, CISO Series — largest CISO podcast network; runs free local CISO meetups.
Steve Wilson · Co‑chair, OWASP GenAI; launched the Agentic Top 10 — the risk list CISOs will map our policies to.
Ken Huang · Creator of CSA MAESTRO, CSA Fellow — writes on “default‑deny at autonomy boundaries” — our model.
Sandy Dunn · CISO; OWASP GenAI core member — leads the OWASP LLM governance checklist — a buyer’s voice.
Rock Lambros · Former CISO; co‑lead, OWASP LLM Top 10 2026 — talks about “agency versus authorization” — our pitch in one line.
No funding yet · product build ≈12 months after raise
- 01Agent‑incident teardowns, monthlyA real coding‑agent breach, and where a block would have stopped it.
- 02Public agent‑action policy templateMapped to the OWASP Agentic Top 10. Free tool comes after funding.
- 03RSAC 2027 talksubmit by 9 OctThen BSidesSF and OWASP chapters. Talks, not booths.
- 0420–30 founder calls → a list of interested companiesWith CISOs and AppSec leads. Design‑partner letters come after funding.
- 05One podcast or webinar with a known voiceShare research, never a pitch.
Wrapbox go‑to‑market research, Sep 2026 · ETH Zurich & Snyk (Invariant Labs, Jun 2025) · Upstarts Media (Keycard) · DXC / 7AI · Canalys (Omdia) 2025 channel forecast · RSAC 2027 call for papers · OWASP GenAI Security Project · each voice’s own 2025–2026 posts, talks and episodes.
$560k builds the product.
Profit in year 4.
- People₹140L
- Product & infra₹30L
- Contingency₹21.5L
- Trust & compliance₹18L
- Setup & admin₹17L
- Go‑to‑market₹10L
Build year. Five people build the product for $269k — under half the raise — with $291k still in the bank at year end.
Three $25k pilots; two convert. The $3M seed lands after the pilots and the team grows to 14.
Ten customers, $800k ARR, a team of 21.
First profitable year: +$288k on $2.4M ARR, 27 people.
49 customers, $5.9M ARR — still under 0.5% of the $1.3B market — with 36 people.
Three founders at ₹40L each and two first hires — GTM and deployment — from month 6. Salaries are 59% of year one.
Cloud, tools and laptops for the five people building the product.
A 10% buffer on every line.
SOC 2 Type I and security testing in year one — the product sells on trust.
Legal, company setup, accounting and a coworking office.
Travel and talks for the pre‑funding motion.
A plan, not a promise · the $3M seed sits below the $7M median of 18 comparable security seeds · cash never drops below $291k
Wrapbox financial model, Sep 2026 · founder inputs · Carta 2025 · Crunchbase News, Jul 2026 · Glassdoor & PayScale, Bengaluru 2025–26 · company releases for 18 comparable seed rounds · ₹88 per US$.
One year to build.
Five to $5.9M.
Now, before funding. Research, the public policy template and 20–30 CISO calls — while we raise the $560k pre‑seed.
A full year to build, on purpose. Five people, SOC 2 Type I, 3–5 design partners giving feedback. Nothing is sold yet.
Launch. Three $25k paid pilots, SOC 2 Type II, and the $3M seed after the pilots. The team grows to 14.
Grow. Ten customers and $800k ARR; we list on AWS and Azure Marketplace. 21 people.
First profitable year. 24 customers, $2.4M ARR; resellers and MSSPs join. 27 people.
Scale. 49 customers, $5.9M ARR, 36 people — still under 0.5% of the market.
Realistic on purpose · the product ships after a full year of building, not in weeks · funding follows the pilots
Wrapbox financial model & go‑to‑market plan, Sep 2026.
I built the controls for when AI goes wrong.
Wrapbox stops it before.
7 years shipping AI and enterprise products — designing the override, escalation and audit screens people use when AI goes wrong.
- UHG / OptumAI benefits and claims assistant: 10,000+ monthly users, 30% fewer support calls. Wildfire AQI Care: an 8‑agent system for 16,000+ members.
- VerizonOps dashboard for stuck orders: 50% faster throughput.
- HDFC BankMoney‑transfer flow: successful transfers up 40%.
- LenskartConversational‑AI builder that teams run without engineers.
9 papers, published and accepted, on agent oversight, escalation and human override — including CAMOR and TCA‑AI.
First hires, month 6: GTM manager · deployment engineer.