AI Threat Watch

Case lovable-rls-cve-2025-48757Model & data leakSweden

Lovable-built apps leak user data: 303 endpoints across 170 projects lacked row-level security

Researcher Matt Palmer found Lovable-generated Supabase backends shipped without working row-level security, letting anyone read or write tables; a scan of 1,645 apps found 170 exposing emails, phones, payment data and API keys.

HighAttack on AIStatus: UnresolvedVerified case file

Open in the interactive mapCite this case

Reported
May 29, 2025
Location
Sweden
Target
Apps generated by the Lovable AI app builder
Severity
High
Triggered by
Matt Palmer (researcher); exposure caused by generated-code defaults
Actor type
Security researchers
Reported loss
None reported
CVE
CVE-2025-48757

What happened

Researcher Matt Palmer found Lovable-generated Supabase backends shipped without working row-level security, letting anyone read or write tables; a scan of 1,645 apps found 170 exposing emails, phones, payment data and API keys.

Impact: 170 of 1,645 scanned apps (303 endpoints) exposed usernames, emails, phone numbers, subscription and payment status and third-party API keys; public exploit reportedly revealed debt amounts and home addresses

Who triggered it

Actor: Matt Palmer (researcher); exposure caused by generated-code defaults

Attribution: Discovered and disclosed by researcher Matt Palmer; Lovable disputed the CVE, arguing customers are responsible for securing their own data

Motive: Security research and coordinated disclosure

How it happened

Attack vector: Direct queries to Supabase REST endpoints using the public anon key embedded in generated front ends, bypassing missing or ineffective RLS policies

Root cause: AI-generated backends defaulted to no effective row-level security and Lovable's later scanner only checked that a policy existed, not that it blocked unauthorized access

Timeline

  1. Palmer reports the flaw to Lovable with proof-of-concept examples
  2. A Palantir engineer publicly posts an exploit; Palmer starts a 45-day disclosure window
  3. Lovable 2.0 ships with a security scan feature
  4. CVE-2025-48757 published; Lovable marks it disputed

Response

Lovable added a security review tool and API-key warnings but disputed the CVE and published no post-mortem; researchers criticised the scanner as giving a false sense of security

Current status: Unresolved

Sources (open in a new tab)

Cite this case

Lovable-built apps leak user data: 303 endpoints across 170 projects lacked row-level security. AI Threat Watch, case lovable-rls-cve-2025-48757, May 29, 2025. https://aithreat.io/cases/lovable-rls-cve-2025-48757/

Permanent link: https://aithreat.io/cases/lovable-rls-cve-2025-48757/

This case file in other languages: EnglishFrançaisEspañolالعربية中文