← Back to Stories

Stories · TMEGS

Linden Lab triages abuse reports 6x faster with AI content moderation on Claude

Mactores rolled out Claude at Linden Lab, the company behind Second Life, supporting AI content moderation for its Trust and Safety analysts and day-to-day development for the engineers who build the platform.
  • AI Agents for Apps
  • TMEGS
  • Claude Enterprise
  • Claude Code
  • Claude Cowork
  • AWS
linden-lab-create-virtual-experiences

Second Life

600K+

monthly active users

~26K

regions

What Mactores delivered for Linden Lab

  • The 95th percentile reflects the slowest reports, so the gain is not an average lifted by quick ones.
  • An engineer counts as weekly active for using Claude Code during the week, not for holding a seat.
  • Up from 42 seats in the pilot.
  • February 2026 to May 2026.

Volume

~1,000

abuse reports reaching Trust and Safety each week

Before

~30 min

p95 handle time per report

Now

~5 min

p95 handle time per report

p95 handle time per report, drawn to scale
Before About 30 minutes This bar is the full track. About 30 minutes before the rollout
Now About 5 minutes One sixth of the track. About 5 minutes today

Each track spans 30 minutes. Hover a bar for the value.

The customer

About Linden Lab

Linden Lab operates Second Life, one of the longest-running virtual worlds built by its own users. The experiences, assets, and scripts that make up the world come from residents, which places engineering and Trust and Safety at the center of how the platform runs.
  • 600K+ monthly active users
  • ~26K regions
  • ~$650M virtual economy a year
  • ~$78M paid to creators annually
Second Life at a glance. Source: Linden Lab

The challenge · 01

The problem

Two groups at Linden Lab stood to gain from Claude, and their work had almost nothing in common.
  • Platform engineering A mature, complex software estate The simulator that runs Second Life regions, the viewer residents use to enter them, and the internal tooling around scripting.
  • Trust and Safety Large volumes of unstructured information Policy, investigations, and sensitive user-generated content, handled alongside governance and creator-support staff.

User-generated content moderation, one report at a time

Trust and Safety carried the sharpest pressure. Every abuse report sent an analyst through the same four steps before anything could happen.
  1. Read the report By hand
  2. Find the relevant policy By hand
  3. Assemble the case context By hand
  4. Decide The analyst’s call

Four lines the rollout could not cross

Any AI rollout also had to respect Second Life’s community.
  • Linden Lab has publicly committed that in-world resident content will not be used to train AI, and residents pay close attention to how the company treats their creations.
  • Moderation content could not drift into engineering workflows.
  • Retention had to stay tightly controlled.
  • Claude could never be the one making an enforcement decision.

The goal was one governed AI foundation that served both groups, with neither team’s needs compromising the other’s.

How Mactores delivered · 02

Okta decides who reaches which Claude

Mactores stood up a single Claude Enterprise organization bound to Linden Lab's Okta identity provider. From there, each person lands on one of two tracks, and the diagram shows what sits on each.
Okta Group membership is the source of truth, so someone who changes teams never inherits the other team's Claude capabilities.

Engineering track engineering Okta groups

Trust and Safety track Trust and Safety Okta groups

Linden Lab's AWS environment Approved MCP services run here and hand back only the minimum results each request needs.
Kept apart groups
workspaces
projects
tools
Platform engineers Onboarded on repository access, Claude Code workflows, and the approved MCP tools.
Claude Code Wired into the source repositories, Jira, Confluence, and documentation the teams already work in.
Claude Cowork + moderation skill The skill packages Linden Lab's triage steps, policy guidance, and human-review rules.

Okta

One identity provider for both teams, with a separate set of groups for each

Group membership is the source of truth, so someone who changes teams never inherits the other team's Claude capabilities.

Engineering track engineering Okta groups

  1. 1 Platform engineers Simulator, viewer, and scripting and tooling teams Onboarded on repository access, Claude Code workflows, and the approved MCP tools.
  2. 2 Claude Code Development, debugging, code understanding, and internal tooling Wired into the source repositories, Jira, Confluence, and documentation the teams already work in.
  3. 3 Engineering MCP tools Approved services, engineering permissions only Allowlisted per track

Kept apart: groups, workspaces, projects, tools

Trust and Safety track Trust and Safety Okta groups

  1. 1 Trust and Safety staff Analysts, governance, and creator support Final decision
  2. 2 Claude Cowork + moderation skill Summarizes the report, surfaces the policy context, prepares the case The skill packages Linden Lab's triage steps, policy guidance, and human-review rules. Zero-retention workspace
  3. 3 Trust and Safety MCP tools Index-scoped read access

Linden Lab's AWS environment Approved MCP services run here and hand back only the minimum results each request needs.

How Linden Lab's Claude organization separates AI content moderation from engineering work. Hover Okta, the AWS tag, or any step marked Details for more.

The work

How Mactores runs it

This is Mactores' AI Agents for Apps work in practice, and it did not stop at launch. Mactores still operates the service with Linden Lab.

First-line support

SSO, provisioning, configuration, and MCP issues on both tracks

Operating runbook

Maintained by Mactores

Handover

Repeatable administration tasks now sit with Linden Lab IT

Monthly review

Adoption, spend, security, and expansion progress, with Linden Lab leadership

Delivery · 03

Two rollout tracks, kept apart on purpose

  1. Security and data-handling review

    Four weeks aligning Claude retention, access, and moderation controls with the commitment on resident content

  2. Claude Code for 24 engineers

    Their feedback fixed access and configuration issues before the track grew

  3. Claude Cowork for 18 staff

    Trust and Safety, governance, and creator-support staff, trained on real moderation work

  4. Company-wide plan ratified

    Linden Lab signs off on taking both tracks beyond the pilot

Access came only after the review, and each objection raised in it left with a concrete answer.

  • Model training

    Deployment settings and contract terms match Linden Lab’s public position on training.

  • Who sees moderation content

    Approved Trust and Safety workflows only, and only for people already authorized to see it.

  • Engineering access

    Engineers’ Okta groups grant no Trust and Safety workspace, project, or tool.

  • Retention

    Moderation data sits in a zero-retention workspace.

In production · 04

The outcome

Claude now takes on the reading, organizing, and policy lookup for each report. Every reviewed piece of Trust and Safety usage stayed inside the access and role boundaries set for the pilot, which is the result the security review was built to guarantee.
Scaling up keeps the tracks separate, because each team's tools, data access, and risk controls differ.
Mactores is also Linden Lab’s partner on an intelligent moderation and creator-support agent, including help for creators working in Linden Scripting Language, which takes Claude further into how Second Life runs every day.

21of 24

engineers across the simulator and viewer squads, active weekly

42 → ~160 seats

pilot seats, then the company-wide plan

What we ship · AI Agents for Apps

Want to know more about Mactores’ expertise in AI agents for sensitive work?

Moderation queues and source code carry very different risks. Mactores designs the identity model, controls, and enablement that let AI agents work in both, in production on AWS.

Know more about Mactores' AI agent expertise

Planning a Claude rollout for several teams at once? Read about Mactores’ generative AI consulting engagements on AWS.