Between 2022 and launch day, GTA VI material leaked through an intrusion that ended with a message in Rockstar's Slack, through a breached analytics vendor and through footage from a development build. The public record of each incident maps onto something a web team can check this week.
Grand Theft Auto VI launched on PS5 and Xbox Series X|S on 19 November 2026. I build websites, not games, and most of what happened inside Rockstar before that is private. The parts that came out through court reporting and company statements still line up with weak spots web teams have too. Where only the attackers say something happened, I mark it as their claim.
| When | What happened | What is confirmed |
|---|---|---|
| Sep 2022 | About 90 clips of early footage posted on GTAForums | Rockstar confirmed a network intrusion |
| Dec 2023 | Trailer 1 leaked a day before its planned release | Rockstar published the real trailer that evening |
| Apr 2026 | ShinyHunters claimed Rockstar data via the analytics vendor Anodot | Rockstar confirmed a third-party breach |
| Aug 2026 | "Cyberleek" leaked footage from what appeared to be an older build | Take-Two subpoenaed Microsoft and Discord |
| Nov 2026 | Cyberleek claimed to release a November 2026 Xbox build on launch day | Unverified, no comment from Rockstar |
September 2022: from a hotel room to the company Slack
On 18 September 2022 someone posted about 90 clips of early GTA VI footage on GTAForums. The poster said they also had source code and a test build, and claimed responsibility for the Uber breach the week before (The Record, PC Gamer). The next day Rockstar confirmed a "network intrusion" and said confidential information, including early development footage, had been downloaded.
The court heard that the attacker, a teenager at the time, was out on bail in a hotel where police had placed him. From that room he posted in Rockstar's internal Slack that he was an attacker and threatened to start releasing source code unless the company contacted him within 24 hours. When police later searched the room, they found an Amazon Fire Stick in the TV and a phone (PC Gamer, The Record).
The hacker claimed on 4chan that he had got in by socially engineering an employee (The Record). Neither Rockstar nor the court has published how he actually got in.
In 2023 he was found unfit to stand trial. A jury found he had committed the acts, and he received an indefinite hospital order (SecurityWeek). He was later found fit for trial, pleaded guilty on 16 November 2026 and will be sentenced at a later date (Aftermath). Rockstar told the court the hack cost $5 million to recover from, plus thousands of hours of staff time (GamesRadar).
Social engineering and MFA fatigue
His guilty plea also covered the Uber hack, and Uber wrote up its incident in far more detail than Rockstar did, so it is the better case study. Uber's security update says the attacker likely bought a contractor's corporate password after malware infected the contractor's personal device. The attacker kept trying to log in, and each attempt sent the contractor a two-factor approval request, until the contractor accepted one. From there the attacker reached other employee accounts with elevated permissions to tools including G-Suite and Slack.
CISA calls this pattern MFA fatigue or push bombing: the attacker keeps sending push notifications until the user approves one by accident or out of annoyance. Its fact sheet on number matching recommends phishing-resistant MFA, with number matching as an interim fix for organizations still on push approvals.
What I would check on a web team:
- SSO, GitHub, the cloud console and the hosting dashboard use security keys or passkeys. Plain push approval without number matching is the first thing to replace.
- Contractors and freelancers get the same MFA policy as employees. Uber's way in was a contractor account.
- Everyone knows that nobody legitimate will ask them to approve a prompt or read out a code.
- A prompt you didn't trigger gets reported, even if you rejected it. CISA's sheet makes the point that MFA runs after the password check, so a denied prompt can mean the password is already compromised.
Chat is an attack surface
Both incidents that September reached Slack. Uber says its attacker posted to a company-wide channel and downloaded some internal messages.
A dev team's chat history holds staging URLs, screenshots of admin panels and the occasional pasted .env. Reading it gives an attacker context, and posting from a colleague's account in #engineering gives them borrowed trust.
- Chat sits behind the same SSO and MFA policy as the rest of your tools, guest accounts included.
- Someone reviews the installed apps and bots, and the tokens they hold, on a schedule.
- Login and export alerts are on if your plan has them, and a named person reads them.
- Secrets go through a password manager. Anything ever pasted into a channel counts as exposed and gets rotated.
December 2023: the trailer that leaked a day early
Trailer 1 was planned for 5 December 2023. On 4 December a low-resolution copy with a crypto watermark spread on X, and Rockstar published the real trailer that evening with a short post pointing people to YouTube (GamesRadar, VGC). Guinness World Records still recognised its 90,421,491 YouTube views in the first 24 hours as the most-viewed video game reveal (Push Square).
Nobody has verified who leaked it, so the useful part for a web team is the response. The finished trailer was ready, and Rockstar put it out the same evening instead of waiting for the planned date. The web equivalent is a launch page or pricing change sitting on a preview deployment. If it leaks at 9 p.m., someone with publishing rights should be able to ship the real one in minutes.
April 2026: the analytics vendor
In April 2026 the group ShinyHunters said it had taken Rockstar data from its Snowflake account through a compromise of Anodot, a third-party analytics vendor, and set a 14 April deadline before it would leak the data (VGC). Rockstar said a "limited amount of non-material company information" was accessed in a third-party breach, with no impact on its players. Snowflake said its own platform was not compromised and put the incident down to a compromise of Anodot. It disabled all user accounts referencing Anodot from connecting. The group claimed 78.6 million records, which Reuters could not immediately verify (Reuters via Insurance Journal).
Snowflake could cut the vendor off by finding accounts that referenced Anodot, and my reading is that this only works when each integration has its own clearly named identity.
A frontend team has more of these than it seems: analytics, an error tracker fed from CI, the hosting platform's GitHub app, a CMS with deploy hooks, AI review tools with repo access. A browser analytics key is meant to be public. Any token that can read data, export it or trigger a deploy belongs in an inventory, and a table in the repo's docs is enough:
| Integration | What it can reach | Credential | Scope | Owner | How to cut it off |
|---|---|---|---|---|---|
| Analytics or BI vendor | Data warehouse | Dedicated service user | One schema, read only | Data | Disable svc_analytics_vendor |
| Error tracker | Release uploads from CI | Auth token in CI secrets | Release uploads only | Frontend | Revoke the token, delete the CI secret |
| Hosting platform | GitHub repositories | GitHub App | Selected repos only | Platform | Uninstall the app or remove repos |
| Headless CMS | Production deploys | Deploy hook URL | One project | Content | Regenerate the hook |
For each row, check whether it can reach more than it needs and whether the credential expires. Fill in the last column before you need it.
August 2026: Cyberleek and the leaked build
In August 2026 a group calling itself Cyberleek spent about a week leaking footage from what appeared to be an older build of GTA VI, and said it had a playable build. Some of the material was posted to a GitHub repository, and Take-Two subpoenaed Microsoft, which owns GitHub, and Discord to try to identify the leaker (Game Developer, Variety). Bloomberg, as cited by Game Developer, reported that Take-Two didn't know who the leaker was or how they got the material, and that after 2022 Rockstar had tightened security, including a full return-to-office mandate.
On launch day Cyberleek claimed to have released a November 2026 Xbox development build (VGC). The claim is unverified, and Rockstar hasn't commented on it.
Subpoenas come after the material is public, and an office mandate decides where people work. Neither controls who can download a build. If you can't say who downloaded build X and when, you can't scope a leak, so internal builds need download logs and expiring links.
A web team's builds leak in quieter ways. The client bundle is public by definition, and Next.js inlines any NEXT_PUBLIC_ variable into the JavaScript sent to the browser at build time (Next.js docs). One wrong prefix ships a secret to every visitor:
// NEXT_PUBLIC_ tells Next.js to hard-code this value into client JavaScript.
// Anyone can now read the token in DevTools.
const exportToken = process.env.NEXT_PUBLIC_ANALYTICS_EXPORT_TOKEN;Modules that read real secrets can import the server-only package. If a Client Component imports such a module, the build fails (Next.js docs):
import "server-only";
export async function getRevenue() {
const res = await fetch("https://warehouse.example.com/query", {
headers: { authorization: `Bearer ${process.env.WAREHOUSE_TOKEN}` },
});
return res.json();
}After each build, a grep over the client output is a cheap smoke test for well-known key prefixes. It only catches formats it knows about:
next build
grep -rnoE "(sk_live_|ghp_|AKIA)[A-Za-z0-9]{8,}" .next/static \
|| echo "No obvious secrets in client chunks"Preview deployments belong behind authentication. Next.js disables browser source maps in production builds by default, and if you turn them on, it serves them next to your JavaScript. If you need them for error tracking, I'd upload them to the tracker and keep them off the public server.
The .env file in the repo
The 2022 threat was to release source code. A code leak hurts far more when the repo also holds live credentials, and a committed .env file is an easy way for them to get there. The .gitignore in Next.js 14's create-next-app template ignored only .env*.local (v14.2.0 template), and by Next.js 15 it ignored .env* (v15.0.0 template). Older projects keep the old pattern unless someone updates it, so a plain .env or .env.production gets picked up by git add .:
# Is any env file tracked?
git ls-files | grep -E '(^|/)\.env'
# Stop tracking it. The file stays on disk.
git rm --cached .env
printf '\n.env*\n!.env.example\n' >> .gitignoreUntracking the file doesn't remove it from history. GitHub's guide on removing sensitive data says the first step for a leaked secret is to revoke or rotate it, and that rotation may be enough without rewriting history at all.
To stop the next one, GitHub's push protection blocks pushes that contain secrets. The per-user version is on by default and covers your pushes to public repositories. For a private repository, an admin has to enable it at the repository or organization level, and that requires GitHub Secret Protection. Locally, gitleaks can run as a pre-commit hook:
repos:
- repo: https://github.com/gitleaks/gitleaks
rev: v8.24.2 # run `pre-commit autoupdate` to get the latest
hooks:
- id: gitleaksRun gitleaks git -v once on the existing history too, because the hook only checks new commits.
Incident response, borrowed from the record
Rockstar confirmed the 2022 intrusion the day after the footage appeared, and by then Take-Two had already started sending takedown notices for the footage (PC Gamer). Uber's update lists containment steps that make a decent first-hour template: it disabled affected internal tools, rotated keys to many internal services, locked its codebase against new changes and required employees to re-authenticate when access came back.
My version for a small web team, written down before it is needed:
- Revoke sessions in SSO and chat, and disable the accounts and integration tokens involved.
- Rotate every secret those identities could read, database and payment keys first.
- Lock the main branch and pause deploys.
- Export logs from SSO, chat, GitHub and the cloud provider before retention deletes them.
- Bring people back with fresh logins and re-enrolled MFA.
- Publish a short statement of what you know, and update it.
I'd stress step 4. In August, according to the Bloomberg report, Take-Two still didn't know how Cyberleek got the material. I can't say what logs Rockstar had, but on a web team, logs exported early are what make that question answerable.
The short checklist
- Security keys or passkeys on SSO, GitHub, cloud and hosting, for contractors too
- Chat behind SSO, with apps and their tokens reviewed on a schedule
- An inventory of integrations that can read, export or deploy, each with a way to cut it off
- No secrets behind
NEXT_PUBLIC_, andserver-onlyon modules that read them -
.env*in.gitignore, push protection on, and one gitleaks scan of the history - A written first-hour runbook, including who can publish a leaked launch early