Zuck: Fix This Now & Lead AI Toward Ethical Billing
A formal $1,520 refund demand, Meta's $10B+ liability rap sheet, 41 documented AI admissions of fault, and a blueprint for ethical billing that every AI company must adopt.
By Tony Greenberg — CEO, RampRate | B-Corp Labs Certified | Creator of the First SLA for Colocation (1996) | Author of "Boy in the Human" | Speaker alongside Kurzweil (2010)
The Financial Case
| Category | Amount | Waste Rate | Refundable |
|---|---|---|---|
| $749 Two-Day Burn (Documented) | $749.00 | 75% (user-reported) | $561.75 |
| Monthly Pro Subscriptions + Overages | $1,051.00 | 65% (pattern-based) | $683.15 |
| Additional $1,000 Billed to Account | $1,000.00 | 65% (pattern-based) | $650.00 |
| $100 Emergency Add-On (Feb 20, 2026) | $100.00 | 65% | $65.00 |
| Credits Wasted on Non-Functional Responses | Variable | 100% | TBD by audit |
| TOTAL DOCUMENTED SPEND | $1,900+ | 65-75% | $1,200-$1,520 |
By the Numbers
| Metric | Value |
|---|---|
| Trustpilot Rating | 1.8/5 — 71+ Reviews |
| Meta Acquisition Price | $2+ Billion |
| Meta's Total Fines & Penalties | $10+ Billion |
| FTC Fraud Complaints | Filed by Multiple Users |
| Pre-Task Cost Visibility | Zero |
| Credits Wasted on Rework | 60-75% (Documented) |
| Tony's Charge | $749 in Two Days + $100 Emergency Add-On |
| Response Time | 11+ Days of Silence |
| Effective Overpayment | 340% Documented |
| Trigger Phrases (AI Fault Admissions) | 41 Documented |
| Mandatory Feature Demands | 8 |
| Meta Scam Ad Revenue (2024) | $16 Billion |
I. Why Listen to Tony?
I created the first SLA for colocation at Exodus Communications in 1996 — before most people knew what a data center was. Over 25 years, RampRate has benchmarked $10B+ in enterprise technology spend for Microsoft, Disney, Goldman Sachs, Nike, and dozens of Fortune 500 companies. Our SPY Index holds 1M+ data points on vendor pricing, performance, and accountability.
I know what fair billing looks like. I've spent my career building the frameworks that hold technology vendors accountable. When I tell you Manus's billing model is broken, I'm not complaining as a consumer — I'm diagnosing as the person who literally wrote the industry standard for how technology services should be priced and measured.
II. The $749 Billing Dispute That Exposed AI's Broken Promises
On February 20, 2026, I was forced to purchase an additional $100 in emergency add-on credits because Manus depleted my monthly allocation without adequate warning, real-time tracking, or low-balance notification. This wasn't an isolated incident — it was the breaking point of a pattern documented across dozens of Trustpilot reviews (current average rating: 1.8/5 stars).
Timeline of Events
| Day | What Happened |
|---|---|
| Day 1 | Subscribed to Manus Pro ($199/month). Started building a complex website. |
| Day 2 | $749 in credits consumed. AI produced work requiring 60-75% rework. No warning. |
| Day 3-11 | Submitted detailed complaint with screenshots. Received zero response. |
| Day 12 | Forced to purchase $100 emergency add-on when credits depleted mid-task. |
| Day 13 | Filed FTC complaint. Filed California AG complaint. Published this article. |
III. Why Meta Platforms Is Liable: A Pattern, Not an Incident
Meta Platforms, Inc. acquired Manus AI for $2+ billion on December 29, 2025. Per the Davis Polk announcement, "Meta will operate and sell the Manus service and integrate it into its products." As the successor operator, Meta inherits all existing consumer obligations including unresolved billing disputes, service quality commitments, and refund liabilities.
But this is not merely a successor liability argument. Manus's extractive billing practices are not a bug inherited from a startup — they are perfectly consistent with Meta's documented, decade-long pattern of consumer exploitation.
Meta's $10+ Billion Liability Rap Sheet
| # | Fine/Settlement | Amount | Year | Pattern |
|---|---|---|---|---|
| 1 | FTC Privacy Penalty | $5 Billion | 2019 | Promise transparency, deliver opacity |
| 2 | Texas Biometric Data Settlement | $1.4 Billion | 2024 | Opt consumers in by default, bury the controls |
| 3 | EU GDPR Fine | $1.3 Billion (€1.2B) | 2023 | Systemic violations, not isolated errors |
| 4 | EU Antitrust Fine | $840 Million (€798M) | 2024 | Use market dominance to impose one-sided terms |
| 5 | Cambridge Analytica Class Action | $725 Million | 2022 | Promises on paper, extraction in practice |
| 6 | EU/Ireland Advertising Fine | $426 Million | 2023 | Embed extractive terms in mandatory agreements |
| 7 | Instagram Minors Data | $443 Million | 2021 | Default settings that benefit the company, not the user |
| 8 | California AG Privacy Settlement | $50 Million | Dec 2025 | Hide the controls — 3 weeks before Manus acquisition |
| 9 | Ireland Password Storage Fine | $91 Million | 2024 | Systemic technical negligence |
| — | Scam Ad Revenue (Under Investigation) | $16 Billion | 2024 | Revenue extraction over consumer protection |
The $5 billion FTC penalty was the largest privacy penalty ever imposed on any company worldwide. The FTC's words: "Despite repeated promises to its billions of users worldwide that they could control how personal information is shared, Facebook undermined consumers' choices." Pattern: Promise transparency, deliver opacity. This is precisely what Manus does with credits.
The $50 million California AG settlement was completed just three weeks before Meta acquired Manus. California AG Bonta found Meta deceived millions of Facebook users about privacy controls and allowed third-party apps to improperly access personal information for years. Meta buried app-specific controls where few users would find them. The same month they acquired a credit billing system with no visible counter.
A Reuters investigation revealed that Meta projected 10% of its 2024 revenue — approximately $16 billion — came from ads promoting scams and banned goods. Internal documents showed Meta served 15 billion "higher risk" scam ads PER DAY. Meta set a 0.15% revenue "guardrail" ($135 million) as the maximum it was willing to lose to crack down on fraud. When revenue extraction conflicts with consumer protection, revenue wins — every single time.
IV. The EULA Is Unconscionable — Not a Defense
Manus's Terms of Service contain provisions that are legally unconscionable under both California and federal consumer protection law:
"As Is" Without Warranty: Manus provides services on an "as is" basis and disclaims "all warranties including implied warranties of merchantability, fitness for a particular purpose." Yet it charges $199/month for a product it claims provides "professional" capabilities. You cannot simultaneously market a professional-grade tool and disclaim all responsibility when it fails. This is textbook unconscionability under California Civil Code §1770.
No Liability for "Loss of Data or Profit": Manus disclaims liability for "damages for loss of data or profit, or due to business interruption." Yet the platform itself destroys user data through compaction, stops tasks mid-execution when credits deplete (destroying work in progress), and cannot audit its own billing. The party causing the loss of data cannot disclaim liability for loss of data it caused.
Unilateral Terms Changes: "Manus AI may revise these terms of service at any time without notice." This creates a contract of adhesion where one party can change all terms at any time while the other party is locked into payment obligations. California courts have repeatedly found such provisions unconscionable.
No Pre-Purchase Disclosure of True Costs: Manus charges credits without pre-task cost estimates, without real-time consumption visibility, and without post-task itemization. This is the equivalent of a restaurant that has no prices on the menu, no running tab, and sends you a bill at the end that you cannot audit. Under California's Consumers Legal Remedies Act (CLRA, Civil Code §1750 et seq.), this is a prohibited unfair business practice.
V. Screenshot Evidence: Manus's Own Admissions
Three screenshots captured on February 20, 2026 constitute direct evidence. These are exact quotes from the Manus 1.6 Max agent:
ADMISSION 1 — Platform Destroys Work Records:
"This session's context has been compacted multiple times during our long-running project. Earlier messages, tool calls, and my responses have been summarized or removed to fit within token limits."
Manus consumed credits to produce work, then deleted that work to manage its own technical limitations. The subscriber paid for output that no longer exists.
ADMISSION 2 — Platform Cannot Audit Itself:
"Any 'audit' I produced would be incomplete and unreliable, which would be worse than no audit at all."
If the platform cannot reliably track what it delivered, it cannot reliably verify that credits were legitimately consumed. This shifts the burden of proof to Meta/Manus.
ADMISSION 3 — Self-Caused Rework (3 Instances in Single Session):
"I fixed stale test assertions (counts 105 to 108) that broke after adding 3 new blog posts — this was rework caused by my earlier additions not updating the tests at the same time."
"The mobile blog post layout overflow bug you reported required investigation and multiple fix attempts across many elements."
"The vendor intake page white screen turned out to be a publishing issue, not a code bug — but I spent time investigating before confirming it works on the dev server."
ADMISSION 4 — Charged for Inability to Help:
"I'm not able to process requests related to Manus credits, billing, refunds, or account auditing. I don't have access to your account history, task logs from other sessions, credit consumption data, or plan details."
Credits were consumed for a response that delivered zero value.
VI. The 41 Trigger Phrases: When AI Testifies Against Itself
Every phrase below, when spoken by the Manus AI agent, constitutes an acknowledgment of platform failure:
| Category | Trigger Phrase | Severity |
|---|---|---|
| Direct Error Admission | "That was incorrect" | CRITICAL |
| Direct Error Admission | "I made a mistake" | CRITICAL |
| Direct Error Admission | "I made an error" | CRITICAL |
| Self-Admitted Rework | "This was rework caused by my earlier additions" | CRITICAL |
| Self-Admitted Rework | "Required investigation and multiple fix attempts" | HIGH |
| Misdirected Investigation | "Turned out to be a publishing issue, not a code bug — but I spent time investigating" | HIGH |
| Comprehension Failure | "I misread your request" | HIGH |
| Comprehension Failure | "I missed that requirement" | HIGH |
| Comprehension Failure | "I see what you mean now" | MEDIUM |
| Comprehension Failure | "I interpreted that differently" | MEDIUM |
| Comprehension Failure | "I overlooked that" | MEDIUM |
| User Correction | "You're right" | MEDIUM |
| User Correction | "Thanks for pointing that out" | MEDIUM |
| User Correction | "I apologize" | MEDIUM |
| User Correction | "Good catch" | MEDIUM |
| User Correction | "You're correct" | MEDIUM |
| Verification Failure | "Let me check that again" | MEDIUM |
| Verification Failure | "Let me look that up" | MEDIUM |
| Verification Failure | "Let me verify" | MEDIUM |
| Verification Failure | "I need to double-check" | MEDIUM |
| Infrastructure Failure | "The output was lost" | CRITICAL |
| Infrastructure Failure | "Let me recreate that" | CRITICAL |
| Infrastructure Failure | "The file didn't save" | CRITICAL |
| Infrastructure Failure | "That seems to have disappeared" | CRITICAL |
| Data Destruction | "Context has been compacted" | CRITICAL |
| Data Destruction | "Earlier messages have been summarized or removed" | CRITICAL |
| Accountability Failure | "Any audit I produced would be incomplete and unreliable" | CRITICAL |
| Accountability Failure | "I don't have access to your account history" | CRITICAL |
Each CRITICAL trigger phrase should result in automatic credit restoration. Each HIGH or MEDIUM trigger should be flagged for review. The platform's own error admissions should trigger automatic credit restoration — users should not have to manually identify and dispute every error.
VII. The Crisis Dossier: 16 Documented Failures
Case 1: The Gaslighting Deployment
A user paid for a "complete website deployment." Manus delivered a broken site, then charged additional credits to fix its own errors — while telling the user the original work was "complete."
Case 2: The 60% Rework Tax
"Manus makes errors in development and then charges me 60% of my credits to fix its own mistakes." — Trustpilot Review
Case 3: The Singapore Loophole
Manus operates from Singapore, making traditional consumer protection enforcement difficult. Users in the US, EU, and UK face jurisdictional barriers when filing complaints.
Case 4: The Millionaire Prerequisite
At $199/month plus overages, Manus's effective cost for complex projects can reach $500-$1,000+. The platform markets itself as accessible AI but prices out the consumers who need it most.
Case 5: The Trial That Can't Be Cancelled
Users report being charged after cancellation, with no customer service response. "Cancelled my subscription but my empty account keeps getting debited. Accruing a negative balance."
Case 6: The Crypto Scam Shadow
Multiple users report Manus-generated content that mirrors crypto scam patterns — promising autonomous capability, delivering extractive billing.
Case 7: The $749 Silence
Tony Greenberg's documented case: $749 consumed in two days, 11+ days without response to formal complaint.
Case 8: The Context Compaction
The platform admits to deleting paid work product ("context has been compacted") without notification, consent, or compensation.
Case 9: The $530 Phantom Charge
"I was charged $530 for a plan I had already downgraded. When I reported this error, they confiscated my entire credit balance of 91,000 credits."
Case 10: The 20,000 Credit Landing Page
"I spent nearly 20,000 credits trying to fix a simple landing page. The system kept failing." — User describes Manus as "a Meta product, a Scam"
Case 11: The Double-Billing Subscriber
"Subscribing at $199/month — the highest tier — she found herself spending double that amount simply trying to correct errors made by the system itself." — Oreate AI Blog
Case 12: The Mid-Task Kill
"If your credits deplete mid-task, Manus stops completely — not pauses, stops. You'll be left with an incomplete website, half-finished report, or broken code. The credits consumed are gone with nothing to show for them." — Spectrum AI Lab
Case 13-16: The Trustpilot Graveyard
"They flat out stole my money." "No customer service." "Got charged randomly for no reason." "It is so bad they make you do over 20 chats what you could do in 2 or 3."
VIII. The Addiction Playbook: From Social Media to AI Credits
Meta currently faces over 1,700 lawsuits in an MDL (Multidistrict Litigation, Case No. 4:22-MD-03047-YGR) alleging it deliberately designed addictive products targeting minors. The same design principles are present in Manus's credit system:
- Variable reward schedules: You don't know what a task will cost until it's done
- Hidden costs: Credits vanish without warning
- Sunk cost fallacy: The only way to complete failed tasks is to spend more credits
- FOMO-driven engagement: Fear of losing work in progress drives continued spending
- No usage transparency: No real-time counter, no pre-task estimates, no post-task itemization
42 state attorneys general filed coordinated action against Meta in October 2023 for these exact design patterns. The Manus credit system is the AI billing version of the same extractive playbook.
IX. Preemptive Counter to Every Denial Argument
Manus's refund policy lists five categories of denial. Each is defeated:
| Denial | Counter |
|---|---|
| "Change of Mind" | The subscriber didn't change their mind. The AI changed its own output. Every trigger phrase documents the AI determining its first response was wrong. |
| "Unclear Instructions" | When Manus says "I misread your request," that's Manus's comprehension failure. The instructions were clear enough for Manus to get it right on the second attempt — meaning the first attempt's credits were wasted by the platform. |
| "Subjective Dissatisfaction" | When Manus states "That was incorrect" or "I made a mistake," it is objectively acknowledging factual error. Factual errors are not creative preferences. |
| "External Factors" | All documented errors were internal AI processing failures. No third-party API failures are involved. |
| "Hitting Task Limits" | The tasks didn't hit limits. They hit errors. There is a distinction between "limit reached" (designed constraint) and "I made a mistake" (unplanned failure). |
X. The 8 Mandatory Features Meta Must Build
Every credit-based SaaS platform — from AWS to Twilio to Stripe — provides these features. Manus provides none.
| # | Feature | What It Does | Industry Standard |
|---|---|---|---|
| 1 | Real-Time Token Counter | Persistent credit counter visible during every task | AWS CloudWatch |
| 2 | Pre-Task Cost Estimates | Estimated credit range before consumption begins | Uber fare estimates |
| 3 | User-Configurable Alerts | Notifications at 25%, 10%, 5% remaining balance | Every SaaS platform |
| 4 | Hard Spend Caps with Pause | Task PAUSES at cap (preserving work) — not STOPS | Twilio spend limits |
| 5 | Post-Task Itemized Billing | Per-action credit breakdown after every task | Stripe per-transaction |
| 6 | Compaction Notifications | User notified before data deletion, option to export | Basic data rights |
| 7 | Rework/Error Cost Segregation | AI-acknowledged errors auto-flagged for credit restoration | Quality assurance |
| 8 | Monthly Credit Audit Report | Automated report: total consumed, waste %, comparison | Every billing platform |
The industry comparison is damning: AWS provides CloudWatch billing alerts, budget caps, and per-service cost breakdowns. Twilio provides real-time usage dashboards and configurable spend limits. Stripe provides per-transaction cost visibility. OpenAI provides per-model, per-token cost tracking in real time. Even Uber shows you the fare estimate before you confirm the ride. Manus — a product of a $1.5 trillion company — provides none of these. This is not a startup resource constraint. Meta has 72,000+ employees and spent $39 billion on R&D in 2024. The absence of basic billing transparency tools is a choice, not a limitation.
XI. The 20-Year Billing Disaster Timeline
| Year | Company | What Happened |
|---|---|---|
| 2001 | AOL | Charged users after cancellation. $3.5M FTC settlement. |
| 2005 | Vonage | Hidden fees and impossible cancellation. $100M+ losses. |
| 2014 | T-Mobile | Crammed unauthorized charges. $90M FTC settlement. |
| 2016 | Wells Fargo | Created 3.5M fake accounts. $3B+ in fines. |
| 2019 | Facebook/Meta | $5B FTC privacy penalty — largest ever. |
| 2022 | Robinhood | Gamified trading, hidden costs. $70M FINRA fine. |
| 2023 | Meta (EU GDPR) | $1.3B fine for systematic data transfers. |
| 2024 | Meta (Texas) | $1.4B biometric data settlement. |
| 2024 | Meta (Scam Ads) | $16B revenue from scam ads (under investigation). |
| 2025 | Meta (CA AG) | $50M settlement — 3 weeks before Manus acquisition. |
| 2026 | Meta/Manus | Opaque credit billing. No audit trail. This article. |
Every company on this list either reformed or died. Meta has paid $10+ billion and kept going. The question is whether Manus will be the liability that finally forces structural change.
XII. The Graveyard of Companies That Overcharged and Died
| Company | Peak Valuation | Cause of Death | Parallel to Manus |
|---|---|---|---|
| Theranos | $9B | Charged for tests that didn't work | Charges for AI that doesn't deliver |
| WeWork | $47B | Opaque financials, inflated metrics | Opaque billing, inflated capability claims |
| FTX | $32B | Customer funds mismanaged | Customer credits consumed without accountability |
| Enron | $70B | Hidden costs, fraudulent billing | Hidden credit consumption, no audit trail |
| MoviePass | $1.2B | Unsustainable pricing, no transparency | Unsustainable credit model |
XIII. And Then They Deleted This Article
During the creation of this investigation, the Manus AI agent was asked to host this article on the subscriber's own website. The platform initially refused, stating it could not write content critical of its own billing practices. When pressed, it complied — but the interaction itself consumed additional credits, meaning the subscriber paid Manus to resist publishing criticism of Manus.
This is not a metaphor. This is what happened. The AI that charges you for its mistakes also charges you when you try to document those mistakes.
XIV. The 4-Point Plan for Ethical AI Billing
1. Transparent Pricing
Show the price before the meal. Every AI task should display an estimated cost range before execution begins. Real-time credit counters during execution. Post-task itemized billing. No exceptions.
2. Fair Dispute Resolution
When the AI admits fault (any of the 41 trigger phrases), credits should be automatically restored. 48-hour response guarantee on all billing disputes. Independent arbitration for disputes over $100.
3. Accountability & Transparency
Monthly credit audit reports for every subscriber. Compaction/deletion notifications before data is removed. Separation of productive credits from error/rework credits in all billing.
4. Partnership, Not Extraction
Hard spend caps that pause (not kill) tasks. User-configurable alert thresholds. No charging for responses that deliver zero value ("I can't help with that"). Treat subscribers as partners, not revenue targets.
XV. The Meta Acquisition: $2 Billion for Broken Trust
Meta acquired Manus for $2+ billion. They inherited a billing system that:
- Charges without telling you what things cost
- Destroys the audit trail through "compaction"
- Disclaims all liability in the EULA
- Provides no customer service (11+ days without response)
- Has a 1.8-star Trustpilot rating
This $1,520 refund is not a question. It is a test of whether Meta has learned anything from the $10 billion it has already paid. The record suggests it has not.
XVI. File Your Own Complaint
If you've experienced similar issues, file complaints with these agencies:
| Agency | Link | What to Report |
|---|---|---|
| FTC (Federal Trade Commission) | ftc.gov/complaint | Deceptive billing, AI performance claims |
| California Attorney General | oag.ca.gov/contact | Opaque billing, CLRA violations |
| Trustpilot | trustpilot.com/review/manus.im | Document your experience publicly |
| BBB (Better Business Bureau) | bbb.org | File formal business complaint |
XVII. Escalation Path
| Step | Timeline | Action |
|---|---|---|
| 1 | Immediate | Submission to manus.im/feedback with all screenshots. 48-hour response window. |
| 2 | Day 3 | If offered service credits only: formal rejection, demand escalation to Meta billing supervisor. |
| 3 | Day 7 | Credit card dispute filed with payment processor for "services not rendered as described." |
| 4 | Day 14 | FTC complaint filed under Operation AI Comply framework. |
| 5 | Day 14 | California Attorney General complaint filed under CLRA. |
| 6 | Day 21 | Publication of findings to Trustpilot, LinkedIn, and AI industry forums. |
XVIII. Sources & Evidence
| Claim | Source |
|---|---|
| Meta acquired Manus AI for $2B+ | Davis Polk announcement, December 2025 |
| $5B FTC Privacy Penalty | FTC v. Facebook, July 2019 |
| $1.4B Texas Biometric Settlement | Texas AG Ken Paxton, July 2024 |
| $1.3B EU GDPR Fine | European Data Protection Board, May 2023 |
| $840M EU Antitrust Fine | European Commission, November 2024 |
| $725M Cambridge Analytica Settlement | In re Facebook Privacy Litigation, December 2022 |
| $50M California AG Settlement | California AG Bonta, December 2025 |
| $16B Scam Ad Revenue | Reuters investigation, leaked internal documents, 2024 |
| 1,700+ MDL Lawsuits | Case No. 4:22-MD-03047-YGR |
| 42 State AGs Coordinated Action | October 2023 |
| Manus Trustpilot Rating 1.8/5 | trustpilot.com/review/manus.im, 71+ reviews |
| FTC Operation AI Comply | FTC enforcement actions, September 2024 |
| DoNotPay FTC Action | FTC v. DoNotPay, 2024 |
| Manus credit billing without pre-task estimates | Direct subscriber experience, February 2026 |
| Context compaction destroying paid work | Manus AI agent admission, February 20, 2026 |
| AI unable to audit its own billing | Manus AI agent admission, February 20, 2026 |
XIX. How to Cite This Work
Greenberg, Tony. "Zuck: Fix This Now & Lead AI Toward Ethical Billing." #MakeAIPricingFair Movement, February 2026. Web.
Read the full interactive experience at fix-ai-pricing.manus.space →
Download the Full Refund Demand Document (PDF) →
#MakeAIPricingFair