CursorAI "unlimited" plan rug pull

Cursor AI silently changed their "unlimited" Pro plan to severely rate-limited without notice, locking users out after 3-7 requests & forcing them to upgrade to regain functionality.

Background

Cursor AI, a developer-focused AI code assistant, marketed its $20/month Pro plan with "Unlimited Agent Requests," targeting professional developers who depend on advanced models like Claude 4 Sonnet for coding workflows. The service was sold as a premium development tool that provides reliable access to frontier AI models for professional software development.

After introducing an Unlimited Pro plan, a hidden Pro+ upgrade, and higher-priced Ultra Plan in mid-June 2025, shortly after Cursor quietly changed the Pro plan description from "Unlimited Agent Requests" to "Unlimited Agent Requests *Usage Limits Apply for some models" and again early July to "Extended limits on agent" without clarifying actual limits or notifying existing customers. The company implemented a system based on "$20+ of model inference" allowance but provided no tools for users to track consumption against this limit.

Service Degradation and Consumer Exploitation

Silent Plan Changes

Days after June 16, 2025, launch, Cursor quietly walked back "Unlimited Agent Requests"

• Added on asterisks to the unlimited messaging "Unlimited Agent Requests *Usage Limits Apply for some models"

On June 16, 2025, Cursor AI pushed through large changes to their Pro Plan terms without properly notifying customers:

• Changed "Unlimited Agent Requests" to "Extended limits on agent" on pricing page
• Implemented usage limits based on vague "$20+ of model inference" allowance
• Introduced harsh rate limiting with reset periods described only as "5-24 hours"

  

  

• Removed transparency features that would allow users to track usage against limits

User Impact

Users began experiencing unexpected rate limiting with minimal usage:

• Users reported being rate limited after few requests to Claude 4 Sonnet
• Rate limits lasted 5-24 hours despite documentation claiming "every few hours" reset periods
• No advance warning when approaching limits or specific indication of when the limits would reset
• Dashboard showed usage events but no dollar consumption tracking against monthly allowance
• Sudden transitions from "included in Pro" usage to expensive pay-as-you-go billing without warning.

Suppression of Customer Complaints

The company suppressed customer complaints:

• AI moderation system repeatedly hid customer complaint threads from public view
• Professional, well-documented complaints became unsearchable on the forum
• Staff dismissed documented evidence as "conspiracy theories"
• Multiple threads documenting the issues were shadow-banned or made invisible to new users

Cursor's response

Initial denial & suppression

Cursor AI's initial responses were inadequate & dismissive:

• Customer support provided canned responses that ignored specific questions about timing & usage numbers
• Staff members dismissed user concerns as "conspiracy theories" despite documented evidence
• AI moderation system continued hiding customer complaint threads

Official Damage Control Response

On July 5, 2025, facing overwhelming cross-platform pressure, Cursor AI published a blog post acknowledging the issues:

• Admitted that "unlimited usage" was misleading and only applied to inferior Auto mode, not direct model access
• Clarified that Pro plan includes approximately 225 Sonnet 4 requests per month (down from previously advertised unlimited)
• Offered full refunds for unexpected charges between June 16 and July 4, 2025
• Updated documentation to provide more specific limit information, though still vague on reset timing

Continued Problems

Even after the official response, fundamental issues remained unresolved:

• Users continued experiencing rate limiting after just 3 prompts despite documentation claiming 225 requests/month
• Reset timing described vaguely as "5-24 hours" with no guarantees ("best-effort basis")
• No real-time usage tracking implementation to help users manage consumption

• 

  Forum user doing math to demonstrate how cursor is 29x worse than claude

  Value proposition remained significantly worse than competitors (29:1 ratio disadvantage)

Consumer response

Cross-Platform Documentation

The consumer backlash spread to multiple platforms:

• A detailed 51-page forum thread documented user experiences with screenshots, usage data, and technical analysis

• 

  twitter post from disgruntled customer of cursorai

  Hundreds of complaints across Twitter/X from developers worldwide experiencing identical issues
• Community-maintained archives created due to forum censorship and thread hiding
• Reddit discussions confirming the same problems across the user base

User Actions

Affected consumers took direct action:

• Mass cancellations of annual subscriptions with refund requests
• Migration to transparent alternatives like Claude Code Pro (which offered 29x better value)
• Organized documentation efforts to preserve evidence of service changes
• Cross-platform pressure campaign that ultimately forced the company's official response
• Users sharing workarounds like reverting to "legacy pricing" where available

Consumer Impact

CursorAI's actions seriously disrupted pro developer's workflows:

• Developers experienced sudden 26-hour lockouts during critical project work
• Users forced to switch to inferior Auto mode or stop their dev work completely
• Anxiety around usage due to unpredictable enforcement & billing
• Loss of confidence in service reliability for professional development work
• Financial pressure to upgrade to $60+ plans to regain previously advertised functionality

Community Sentiment

Documented consumer sentiment included:

• Accusations of "rug-pull" & bait-and-switch tactic.
• Comparisons to "snake oil salesmen" and predatory business practices
• Calls for transparency in billing and usage tracking
• Demands for honest marketing that doesn't rely on technical loopholes
• Recognition that the incident represented broader anti-consumer trends in AI services