A researcher revealed that the vibe-coding platform Lovable uncovered customers’ chat histories with AI fashions to different customers accessing the platform via an API (software programming interface).
X person @weezerOSINT, reported the publicity in a post on Monday. “I made a Lovable account immediately and was capable of entry one other person’s supply code, database credentials, AI chat histories, and buyer information are all readable by any free account,” the researcher wrote. The put up included a screenshot of one other Lovable person’s challenge code and chats, together with an unresolved ticket for the bug that allegedly brought about the information leak.
In a follow-up dialog with Quick Firm, @weezerOSINT (who didn’t share his actual title) says it took half-hour utilizing xAI’s Grok 4.2 mannequin to conduct the analysis, including that earlier than AI, discovering related exposures would take hours or days.
@weezerOSINT reported the difficulty by way of HackerOne, a cybersecurity firm that runs bug bounty and vulnerability disclosure packages, in early March. On Monday, the researcher confirmed that Lovable initiatives created earlier than November 2025 nonetheless expose the information.
Lovable declined to offer an govt to clarify the scenario, and pointed to its public assertion on X.
Lovable initially said on X that no “information breach” had occurred, and that exposing challenge code was “intentional conduct.” When customers mark their initiatives “public,” the corporate defined, they choose to have their code seen to different customers.
However that didn’t account for the publicity of customers’ chats and prompts with the AI mannequin, which Lovable made accessible for public initiatives till not too long ago.
“We additionally retroactively patched our API so public challenge chats couldn’t be accessed, it doesn’t matter what,” Lovable mentioned in a second, clarifying post on X. “Sadly, in February, whereas unifying permissions in our backend, we by accident re-enabled entry to chats on public initiatives.”
As for @weezerOSINT’s early-March report back to HackerOne, Lovable says the ticket had been closed as a result of its “HackerOne companions” believed that viewing public initiatives’ chats was “the meant conduct.”
As a vibe-coding platform, Lovable treats natural-language prompts used to generate code as a core a part of the constructing course of. The corporate initially believed its group would profit from seeing how different builders used prompts to construct options, features, parts, or database schemas, so chats have been handled as customary challenge metadata.
However the danger of exposing delicate data in these chat histories seems to have outweighed that profit. Lovable says that in December 2025 it made all new initiatives “personal by default” for all customers.
Lovable’s most recent funding round got here in December 2025, when it raised $330 million from CapitalG, Menlo Ventures, Khosla Ventures, and others. After the spherical, the corporate was valued at $6.6 billion, reportedly tripling its valuation in about 5 months.
