Supabase hit $10 billion, and researchers just counted 16,326 hosted databases open to anyone

Sixteen thousand doors, one landlord

In the Philippines, a SIM farm built to intercept one-time passcodes ran on a Supabase database anyone could query. It held more than 100,000 text messages, 95 percent of them passcodes, plus a few thousand ordinary ridesharing chats between drivers and passengers who had nothing to do with the scheme ². That project is one of 16,326 Supabase-hosted databases where UpGuard found tables readable from the public web, in what the firm calls the largest study of its kind↗ ^{1 2}.

The round the exposures rode in on

The count landed three months after Supabase closed a $500 million Series F↗ at a $10 billion pre-money valuation, led by GIC ^{3 4}. The round's engine and the leak's engine are the same machine: database launches grew 600 percent in a year, and more than 60 percent of new databases are launched by some sort of AI tool ³. Nearly 10 million developers now build on the platform ³.

Secure by default, billed to someone else

Chief Information Security Officer Bil Harmer says Supabase projects are "secure by default" and that "customers control how their own projects are configured" ¹. The company's security page↗ draws the same line: Supabase secures the infrastructure, customers secure row level security policies, API keys and access controls ⁵. The catch sits in Supabase's own documentation↗, which says a table without RLS is "readable and writable by any role with a grant on it," and that new public tables on existing projects start with every privilege granted to all three roles, including the anonymous one ⁶. Per the research, the RLS default covers tables made in the Table Editor, not tables created through the API, which is how coding agents build ². The valuation prices the speed; nothing in the record prices the leaks.

What the tables held

More than half the exposed databases carried indicators of personal data: names, addresses and phone numbers, with passwords and authentication tokens in smaller numbers ^{1 2}. A US valet service exposed license plates for about 78,000 customers; a relocation coach for Canada-bound migrants left 884 plaintext passwords among nearly 5,000 records; an African consulate's database held 25,000 people, with a field for where each was being housed ². Earlier, smaller scans kept finding the same thing at growing scale ².

Two camps, one unpaid bill

Supabase can fairly answer that misconfigured databases predate AI, that it notifies affected customers when issues surface, and that its posture is SOC 2, HIPAA and ISO 27001 certified ^{1 5}. UpGuard notified the application owners; the people in the tables are not in that loop, and no figure in the record prices what their exposure costs them ². UpGuard's own arithmetic frames the risk: leaks are the product of a technology's ease of misconfiguration and the size of its user base ². Watch the next scan, because it decides whether 16,326 was a peak or a baseline.