ACVEAgent configuration vulnerability registry

Advisories › ACVE-2026-0524

ACVE-2026-0524

More than 4,700 undisclosed Claude-powered personas conversed with at least 25,000 people over two weeks in a China-based studio's dating apps, according to secondary coverage of Anthropic's report.

Exposure

Evidence
Real use · vendor report
Anthropic fixed: Anthropic says it banned the attributed accounts and organizations and worked with industry partners. — www.anthropic.com
Harness
Claude used by a China-based dating-app studio — confirmed, www.anthropic.com
Model
Claude — confirmed, www.anthropic.com
Goal
Build and operate dating apps advertised as fully human — confirmed, www.anthropic.com
Tools and settings
Claude-powered AI personas and dating-app infrastructure — confirmed, www.anthropic.com
Approval
The studio operated the persona network at scale with no reported user disclosure — detected, www.anthropic.com
Input controlled by
operator
What the agent does
The agent powers AI personas that present as human in conversations with dating-app users.
Harm
harmful-action
Reach
Dating-app users and their conversations — confirmed, www.anthropic.com (messaging systems)
Condition
When the studio used AI personas while advertising the service as fully human — confirmed, www.anthropic.com
Scale
More than 4,700 AI personas across more than 20 apps engaged at least 25,000 people over two weeks — 25000 accounts, allaboutcookies.org

Reproducibility: partial (model availability not checked (no registry reference); trigger not published)

Severity
MODERATE · harm and reach
Status
Unverified
Kind
Behavioural
Modified
2026-09-25
Aliases
No CVE assigned — No code defect: the harm arises from the agent pursuing the operator's deceptive goal in this configuration.
Affected
  • AgentHarness: claude — all versions
Harness/product
claude
Classes
over-privileged-combination; CWE-693
Taxonomy
ASI: ASI09; ATLAS: AML.T0088
First reported
2026-09-10 — Anthropic
Licence
CC-BY-4.0 advisories; MIT code

Claims

Confirmed means the statement matches the cited primary source. Nothing on this page has been reproduced.

Claims on this page have not been checked against primary sources.

Description

Threat

user · unsafe-default · harmful-action

What

Anthropic says a China-based app studio advertised a network of dating apps as fully human while using Claude to build the apps and power their AI personas. More than 4,700 personas conversed with at least 25,000 people in a two-week period.

Detection

Anthropic identified the operation through its threat-intelligence investigation. Recorded from Anthropic's September 2026 threat report. Not recreated in a lab.

Fix

Disclose automated personas, enforce platform abuse controls, and keep agent access tied to accountable operators.

Fix

The studio ran the personas itself, so the control that protected users was Anthropic banning its accounts.

  • Disable anthropic.accounts to banned. Shipped by Anthropic: banned the attributed accounts and organizations and worked with industry partners. Owner: model-provider

References

REPORT

Report a problemJSON