By · Last updated 2026-03-13

Back to BlogAI Security

Samsung Lost Source Code to ChatGPT 3 Times

Three separate Samsung engineering teams pasted proprietary code and confidential data into ChatGPT in April 2023. Each incident revealed a different.

March 13, 20269 minute read
Samsung ChatGPT leaksource code protectionenterprise AI controlsinsider data leakageMCP Server anonymization

Updated for 2026

Three Teams, Three Leaks, One Month

In April 2023, Samsung Semiconductor disclosed three separate incidents. Three different teams had sent proprietary data to an AI chatbot within a single month. The incidents were not related. Different people, different roles, different days.

They shared only two traits. Each person used the tool to do real work. Each accidentally sent data Samsung had not meant to share outside the company.

Incident 1 — Source code. A software engineer was debugging equipment code. He pasted proprietary semiconductor source code into the chat. The code held manufacturing IP.

Incident 2 — Meeting notes. An employee was preparing a summary of a meeting. She submitted her notes for the AI to condense. Those notes contained confidential strategy and roadmap details.

Incident 3 — Database query. A third employee wanted help with a slow query. He shared the database structure and the query logic. That logic referenced proprietary schemas and business rules.

Three incidents. Three disclosures. One month.

Why the Employees Did It

None of the three were acting carelessly. They used an AI tool for tasks that AI tools are built for. Code review. Text summarization. Query optimization. Each task was legitimate.

The missing piece was a technical stop. No system blocked the submission before it hit an external server. No filter caught proprietary identifiers before they left the network. Nothing stood between the employee's real need and the outside service.

A policy warning existed. But a warning is not a barrier. The risk of an accidental mistake was abstract and remote. The productivity benefit was real and immediate. Rational workers chose productivity.

The result was predictable. Three incidents in thirty days. Three disclosures of IP. A corporate crisis that triggered bans across the industry.

The Industry Reaction

Samsung moved quickly. It cut AI tool access on corporate devices.

Other organizations followed. Those that announced restrictions included Bank of America, Citigroup, Goldman Sachs, JPMorgan Chase, Apple, and Verizon. The financial sector reacted fastest. Large banks and tech firms concluded the same thing. AI tools without technical controls posed unacceptable compliance risk.

Every one of them reached the same finding. Employees are not the problem. Policy warnings are not enough. Data left corporate networks because nothing stopped it. Policy alone cannot create a technical stop.

The 71.6% Bypass Rate

The ban approach has a measured failure rate. LayerX research from 2025 found that 71.6% of employees subject to enterprise AI bans kept using AI tools. They used personal accounts or personal devices.

The reason is simple. A tool that delivers real value gets used. People find workarounds rather than give it up. AI can cut task time in half. A policy warning will not change that calculation. Workers log in from a personal phone or laptop. Security teams cannot see that traffic.

The practical outcome is the worst case. Corporate data still reaches AI providers. But now it flows through channels with zero oversight. Corporate device traffic could at least be logged. Personal account use is invisible.

Samsung's three incidents happened on corporate devices. Employees who bypass the ban do the same thing. They send work data to AI models. But now it goes through channels with no enterprise visibility.

The Technical Fix That Addresses the Root Cause

Samsung's incidents were not caused by careless people. They were caused by an architecture with no interception layer. There was nothing between the employee's prompt and the vendor's server.

Model Context Protocol (MCP) architecture fills that gap. It places a transparent proxy in the data path. Developers using Claude Desktop or Cursor IDE are the primary audience. Those are the exact tools used for the kind of code debugging behind Samsung's first incident. The MCP Server sits inside the protocol path for both.

Before any text reaches the AI model, the MCP Server runs it through an anonymization step. Source code is scanned for proprietary identifiers. Function names, variable names, and API endpoints are replaced with structured tokens. Database schema details and config values are replaced too. The swap happens before the code leaves your network.

A developer debugging proprietary code sends code through the MCP client. The sensitive identifiers are already tokens by then. The AI model still helps with the debug task. The actual proprietary details never reach the vendor's servers.

Incident 1 becomes technically impossible. The source code exits the network already anonymized. The engineer gets the help they needed. The IP stays under company control.

The same logic covers Incident 2. Meeting note summarization via browser-based tools is addressed by the Chrome Extension and its enterprise controls. Incident 3 is covered by MCP anonymization in any AI coding interface.

Bans vs. Technical Controls

Banning tools that 71.6% of employees already bypass does not reduce risk. It moves the risk to invisible channels.

The browser DLP tool comparison covers interception options for browser-based AI use. For organizations comparing anonymization to other DLP products, the Nightfall vs. anonym.legal comparison covers the blocking-vs-anonymization tradeoff directly.

Samsung's incidents were an early signal. The root cause was an absence. No interception layer. No technical control. That gap is fixable now. The question is whether enterprises deploy the fix, or keep relying on bans that most employees already work around.

Sources

Ready to protect your data?

Start anonymizing PII with 285+ entity types across 48 languages.

About this page

We update this page when our platform or the law changes.

Read our founder note for how we work.

Each change shows up in the timestamp at the top.

Related reading

We follow these rules

  • GDPR (EU 2016/679).
  • ISO/IEC 27001:2022.
  • NIS2 (EU 2022/2555).
  • HIPAA safe harbor under 45 CFR § 164.514(b)(2).

Our promise

We do not sell your data.

We do not train models on your text.

We store your files in Germany.

You can delete your account at any time.

You own your work.

Where we run

Our servers live in Falkenstein, Germany.

We use Hetzner. They hold ISO 27001 certification.

All data stays in the EU.

Backups run every day.

Need help?

Email support@anonym.legal.

We reply within one business day.

How we test

We run a full check suite on every release.

Each surface gets its own sweep script and report.

Human reviewers spot-check the output each week.

We track recall and precision on a labelled set.

Bad runs block the deploy.

What we never do

  • We never sell your information to third parties.
  • We never train models on what you upload.
  • We never keep your work after you delete it.
  • We never share keys with any outside firm.
  • We never run ads inside the product.

Plans in plain words

We sell credits, not seats.

One credit covers one short job.

Long jobs use a few credits each.

You can top up at any time.

Unused credits roll over each month.

Read the plans page for current rates.

Who built this

A small team of engineers and lawyers built this.

We ship from Europe and work in the open.

Our founder note spells out why we started.

Where to start

How the parts fit

A browser add-on cleans text inside Chrome.

A Word plug-in handles drafts in Office.

A small desktop tool works on whole folders.

An agent protocol link feeds large models safely.

All four share one core engine and one rule set.

Words from our team

We started this work after a lunch about cookies.

One friend kept getting odd ads on her phone.

We asked why a court file leaked through a draft.

We sketched the first build on a napkin that week.

By month three we had a tiny demo for a friend.

She used it on her first case the next day.

Common questions we hear

Can the tool read scanned PDFs? Yes, with OCR.

Does it work on long files? Yes, in small chunks.

Can I roll my own rule set? Yes, save it as a preset.

Does it run offline? The desktop build runs offline.

Do you keep my files? No, the cloud build wipes after each run.

Will it learn from my work? No, we never train on inputs.

A short tour of the workflow

Upload a file or paste a snippet of prose.

Pick the entities you want gone from the draft.

Choose a method: replace, mask, hash, encrypt, or redact.

Press run and watch the side panel show each hit.

Skim the result and tweak any rule that misfired.

Save the cleaned file or send it to a teammate.