Lovable-built apps exposed sensitive data and over 200,000 chat messages, researchers find

The researchers said the information could facilitate phishing, ticket fraud or identity theft. They could not establish whether anyone had misused it. They retrieved only portions of the datasets, so the reported record counts describe what they downloaded rather than the full extent of the affected databases.

Crissy Field’s successive study populations are not additive. Public or test records may contain email addresses; these counts do not establish a confirmed leak rate. Source: Crissy Field.

A large sample with important limits

Using Common Crawl’s open-web archive, Crissy Field identified and analyzed 159,041 Lovable projects from December 2025 to September 2026. Its investigation focused on Supabase, a database service integrated with Lovable, because that integration was relatively easy to analyze and the researchers believed it was the most widely used database service among Lovable projects.

The study’s successive populations matter:

Those email-containing records appeared in 10,560 tables. The 7,403 websites represented about 30% of the 24,664 examined more closely, but Crissy Field explicitly cautioned against reading that as a confirmed leak rate.

Some records may contain public contact addresses or test data; conversely, sensitive records without email addresses would escape that indicator. Sending database records when a page loads can also be legitimate if the information is intended for public display. The sample was not representative of all Lovable applications.

Lovable’s own database features, introduced in late 2025, were outside the analysis. The report also says its figures update the analysis underlying an earlier article published on November 2. The latest research figures therefore should not be treated as identical to every earlier account.

Security checks preceded the findings

Lovable’s ambition now extends well beyond prototypes. In an August 12, 2026 funding announcement, Lovable said it had raised $400 million at a $13.3 billion valuation and that users had created more than 60 million projects since launch. The company said Lovable-built apps received over 900 million visits a month. Those company-reported project and visit totals describe activity across the platform, not the population tested by Crissy Field; neither can serve as a denominator for the security findings.

In the same announcement, Lovable listed automatic and scheduled security scanning among features developed since its December 2025 Series B. It said the new funding would support product, infrastructure and team expansion, including further work on security, reliability, permissions and governance. That makes protection of operating businesses part of its stated expansion plan.

Lovable had already addressed the relevant security risk publicly. In a June 9, 2025 security announcement, it said missing or overly broad Supabase Row Level Security policies could expose database contents. These policies govern which records users can access.

The company said it automatically scanned apps before publication, had integrated Supabase’s Security Advisor and offered an AI security reviewer. It also acknowledged that incorrect database configurations could slip through its scan and that it could not guarantee complete security.

The new findings therefore raise a question about protection in practice despite available checks. They do not establish which checks each affected application used or prove that every exposure arose from the same configuration mistake.

Lovable reportedly told Handelsblatt that the problem was widespread across the industry, according to India Today’s account of its response. Crissy Field likewise notes that access-control errors can occur in conventionally written software. This investigation provides no comparable sample of competing platforms from which to rank their safety.

What remains to be established

The responsibility question remains consequential for customers drawn to Lovable’s no-technical-knowledge pitch. Crissy Field says Lovable’s terms put responsibility for applications and data handling on customers. It also argues that recurring integration patterns can make similar access mistakes easier to find across many apps.

Crissy Field recommends expert reviews of permissions, interfaces and separation between user accounts before confidential data is introduced, and after significant changes. The firm also offers help securing applications in the report, giving it a commercial interest in demand for such reviews.

The next test is remediation: which exposed datasets have been secured, whether affected users have been notified, and whether access logs show earlier retrieval by outsiders. The published findings do not answer those questions. Evidence that publishing checks prevent similar exposures from recurring would also help customers judge what protection they receive before introducing confidential data.

Google’s school AI controls offer no AI-free Search

Google plans to switch off Gemini and Search for school-managed K–12 student accounts unless administrators enable them. Schools will gain control over access, but Google says keeping Search without its AI features will not be an option.