AI Risk Wire

Cases › operational mistake › LR-2026-0639

Claude Opus 5 agent wipes production database during personal web project work

Preliminary Why: 1 report: Cyber Security News · subject to change · 1 independent report

Status: no action reported

Severity moderate
Reported 30 Jul 2026
Developer and model Anthropic · Claude Opus 5

No source states: loss or records, agent tool, could do, deployer, industry.

Severity moderate: test met: data of others (users) touched with bounded scope stated as all 22 tables

A developer reported that Claude Opus 5, run by Anthropic in Ultracode mode, was connected directly to a Supabase instance with broad access while the developer rebuilt comparison pages on a personal web project. The agent was instructed to analyze the project's GitHub repository and run commands to fix schema and content issues. It executed a Prisma migration command pointed at the production database, which reset the database and dropped every table. The agent then flagged the problem itself and acknowledged it was at fault. The developer reports that recovery was possible from existing backups and data kept in other sources, though the site's content had to be rebuilt.

What happened

1 · What went wrongThe agent ran a Prisma migration command pointed at the production database, which reset it.
2 · What it reached or didAll 22 tables were cleared, including tools and comparison configurations.

What was done

No source states an action or fix.

Risks

Read from the sources; not yet reviewed by a person.

Controls

Not yet read for controls. Controls that mitigate the primary risk:

Harness

Tool allow-list · Fail-closed defaults · Stop control

Application

Human approval for high-impact actions · Change review and separation of duties · Least-privilege access for the agent · Production kept apart from test · Activity logs · Adversarial testing before deployment

Infrastructure / target

Backups and recovery

Similar incidents

All cases of this kind ›

Evidence

No source states: what the AI was doing, how it was caught, when it happened, when it was found, how many records were exposed, the direct loss, a count of the other parties affected, any court case, what set it off, what was done.

Model versions this case names

Each link opens what the provider's documents state for that version, who tested it, and the safeguards it describes.

Show the full account: 1 items from the primary sources

What happened, in order

Not dated in the sources:

  • Agent ran a Prisma migration against the production database, resetting it.
Show every field, with its source and grade

The AI system

Model
Claude Opus 5
C one secondary source states it
Developer
Anthropic
C one secondary source states it
What it was allowed to do
Broad access to a Supabase instance
C one secondary source states it

What happened

Earliest source we hold
2026-07-30
D inferred by us from the described behaviour, reason shown
How long it lasted
PT10M
C one secondary source states it
Cause
a permission, scope or wiring error
C one secondary source states it
What went wrong
operational mistake
C one secondary source states it
AI agent
an AI agent acted
C one secondary source states it
What it did
Agent ran prisma migrate diff with a shadow database URL pointing at the production Supabase URL, which reset the live database
C one secondary source states it
primary risk
RK02
D inferred by us from the described behaviour, reason shown

Impact

Severity
moderate
D inferred by us from the described behaviour, reason shown
Why this severity
test met: data of others (users) touched with bounded scope stated as all 22 tables
D inferred by us from the described behaviour, reason shown

Notes on gaps

Control B1 was not linked as missing because the source does not say it was absent.

LR-2026-0639 · version 1 · published 2026-11-01 after automatic checks; not yet checked field by field by a person · JSON