Report ID: 1219 — Source-code-leak / intent-of-codebase analysis
On 18 September 2022, a user calling themselves "teapotuberhacker" posted roughly ninety video files to GTAForums purporting to come from an internal test build of Grand Theft Auto VI (Fanelli, 2022; Bonifacic, 2022). Bloomberg's Jason Schreier confirmed within hours, via Rockstar sources, that the material was authentic, calling it "one of the biggest leaks in video game history" while emphasising that "the footage is early and unfinished, of course" (Schreier in Bonifacic, 2022; Schreier in Fanelli, 2022). Rockstar Games corroborated the breach in a public statement the following morning (Bonifacic, 2022).
The dominant tone of mainstream coverage in the days that followed — across Bloomberg, Eurogamer, GameSpot, Engadget and the wider trade press — was that the leaked clips were not finished mission scenes, scripted cinematics, or vertical-slice demos. They were, in the words of working developers who reshared their own equivalent material in solidarity, "what art looks like for a video game in development" (Margenau in Phillips Kennedy, 2022). Reporters described untextured environments, missing assets, placeholder set-dressing, and a striking emphasis on systems being exercised in isolation: a character walking through a strip club, another riding the "Vice City Metro", a male protagonist firing what looked like an AK-47 at passing police, and the two leads collaborating to rob a diner (Bonifacic, 2022; Fanelli, 2022).
This report does not analyse the leaked footage itself. It analyses what mainstream reporters said about the footage and asks a narrower, more interesting question: what does the composition of that internal test build — heavy on animation, traversal and gunplay tests; light on polished mission content — imply about Rockstar Games' design priorities for GTA VI? The argument advanced here, strictly as inference from public press accounts, is that Rockstar appears to be pursuing a "feel-first" design philosophy: locking down the moment-to-moment kinesthetics of the systemic sandbox before narrative scaffolding is built on top. That order of operations inverts the industry-typical "vertical slice" approach to AAA development and constitutes, if the inference holds, a substantial bet that systemic feel — not scripted set-pieces — will sell the game.
All speculation about specific animation systems, blend trees, locomotion graphs or capture pipelines is clearly labelled as inference and is not drawn from any leaked code, file paths or internal identifiers. The discussion stays at the level of design philosophy that mainstream reporting has already surfaced in print.
The single most consistent characterisation of the leak across mainstream outlets was that the material was early, raw and systems-focused rather than scripted or curated.
GameSpot summarised the trove as "pre-alpha footage and multiple screenshots" and catalogued the contents as a female protagonist walking through a strip club, a male character "shooting what looks to be an AK-47 at passing police cars", and the two leads "working together as they rob a diner" (Fanelli, 2022). Engadget framed the same material as ninety videos "from a test build" and noted that, in line with Schreier's earlier July 2022 reporting, the footage exposed two playable protagonists, a Miami-coded setting, and clips of the second protagonist "riding the 'Vice City Metro'" (Bonifacic, 2022). The vocabulary — "test build", "early and unfinished", "pre-alpha" — recurs almost verbatim across outlets.
Eurogamer's follow-up coverage is the most pertinent for this report because it foregrounds the production-stage reading of the leak rather than the lore reading. The outlet reported that "[o]ver the weekend, Rockstar faced a massive leak that saw numerous screenshots and videos taken from GTA 6's development show up online with missing assets and features" and devoted the body of the article to active developers contextualising what the public had seen (Phillips Kennedy, 2022). Uncharted 4 co-lead designer Kurt Margenau described what was visible as "what art looks like for a video game in development", sharing his own "blockmesh vs art blockmesh vs final art" comparison (Margenau in Phillips Kennedy, 2022). Naughty Dog Central reshared similarly raw early footage of The Last of Us; Remedy's Paul Ehreth shared development clips from Control; Sam Barlow shared early Immortality stills; Cian Maher reshared a Horizon Zero Dawn Thunderjaw prototyped "using Killzone assets because games don't look like the finished product until extremely late in development" (Phillips Kennedy, 2022). The point of that solidarity wave — and the reason it matters here — is that the working-developer consensus, as reported by Eurogamer, was that the leaked clips were structurally normal for a project at that stage: animation, locomotion and gameplay-systems work surfaced ahead of art passes, lighting passes, mission scripting and final environment dressing.
What the press did not describe is equally instructive. Reporters did not surface lengthy scripted cinematics, finished dialogue trees, polished cutscene direction, or anything resembling a curated demo reel of the kind a publisher would assemble for an E3-style showcase. They described systems exercises: walking, riding transit, shooting, robbing, traversal between interiors and exteriors. Bloomberg's framing throughout, via Schreier, emphasised that the build was a working internal artefact, not a marketing artefact — "early and unfinished, of course" (Schreier in Bonifacic, 2022; Fanelli, 2022).
It is reasonable to infer (and labelled as inference) that the ratio of content types in the build — systems and animation tests numerous and prominent; finished narrative content largely absent — reflects what Rockstar's teams were spending their time on at that point in development. That ratio is the central piece of evidence the rest of this report works from.
The dominant production paradigm in modern AAA development is the vertical slice: a short, heavily polished segment of the game — usually one mission or one self-contained area — taken to near-shipping quality early in development to prove out the entire production stack (art, animation, audio, mission scripting, UI, cinematics) and to function as both an internal alignment artefact and an external pitch tool. The vertical slice is, by design, deceptive about the rest of the game: it tells you what the best moment of the finished product is expected to look like, not what the median in-development state of the project actually is.
Mainstream press descriptions of the September 2022 leak imply, by inference, that Rockstar's internal build was structured very differently. Phillips Kennedy's Eurogamer round-up (2022) effectively functions as an industry tutorial on the point: the developers who came forward in solidarity were not saying "this is unusual but normal for Rockstar"; they were saying "this is what every game looks like at this stage behind the vertical slice". Margenau's "blockmesh vs art blockmesh vs final art" comparison, Naughty Dog Central's early Last of Us footage and Maher's note that Horizon Zero Dawn prototyped using Killzone assets all point to the same hidden reality: the systems-and-animation layer underneath the slice exists in every studio, but it is normally never seen outside the building (Phillips Kennedy, 2022).
What is distinctive about Rockstar's case — strictly as press-reported inference — is the composition of the build that was leaked. Across Bloomberg, GameSpot and Engadget, no outlet described a curated vertical-slice mission within the trove. The descriptions are of a kinesthetic playground: animation tests for traversal, gunplay prototypes, shared-character robbery interactions, transit-riding behaviour (Fanelli, 2022; Bonifacic, 2022). If a polished vertical slice existed inside Rockstar in mid-to-late 2022, it was not what the hacker exfiltrated, and it was not what Rockstar's own sources, speaking to Bloomberg, characterised as the state of the project (Schreier in Bonifacic, 2022).
The inference — and it is inference, not fact — is that Rockstar at that point was prioritising systemic depth over showpiece readiness. That is the opposite ordering of priorities from the vertical-slice paradigm. In the vertical slice approach, you make one mission beautiful first and then scale the rest of the game up to its standard. In what the leak's composition implies, you make the verbs of the game feel correct everywhere first, and the missions are written and scripted around verbs that are already locked.
This inversion is not unprecedented in Rockstar's published history. Bloomberg's own reporting on Red Dead Redemption 2, by Jason Schreier (2018) at Kotaku before his move to Bloomberg, repeatedly emphasised the studio's obsession with simulation depth — horse testicles shrinking in cold weather, NPCs reacting to the player's appearance and hygiene, animation transitions taking precedence over animation count. That reporting characterised RDR2 as a project where moment-to-moment realism was not subordinated to mission delivery but, if anything, the other way around (Schreier, 2018). The 2022 leak coverage, taken together, suggests Rockstar is doubling down on that ordering for GTA VI.
If the inference holds — that Rockstar prioritised animation, traversal and gunplay tests early, with mission content built around them rather than the other way round — then the choice is not aesthetic. It is commercial. It is a bet about what actually sells a Grand Theft Auto game.
A vertical-slice-first studio is implicitly betting that marketing moments sell the game: the trailer beat, the showcase mission, the E3 demo that gets reshared as a GIF for years. A feel-first studio is betting that the first ten minutes of unsupervised play sell the game: how it feels to walk, run, draw, aim, drive, crash, get out of the car, climb a wall, take cover, switch protagonists. The two strategies are not mutually exclusive — every AAA game needs both — but the order in which a studio invests is revealing about which one it believes is load-bearing.
The press framing of the 2022 leak is consistent with Rockstar believing the second. Bloomberg's coverage (Schreier in Bonifacic, 2022; Fanelli, 2022) and Eurogamer's developer-solidarity piece (Phillips Kennedy, 2022) both implicitly characterise the leaked build as a playable system rather than a demoable slice. Reporters who saw the clips described actions and behaviours, not story beats: robbing a diner, riding transit, firing weapons at pursuing police. Those are verbs, not narratives.
There is a strong commercial logic to that bet for Grand Theft Auto specifically. The franchise's economic engine — GTA Online's decade-plus monetisation tail on GTA V — does not depend on scripted set-pieces. It depends on the moment-to-moment satisfaction of being inside the world: driving feels good, shooting feels good, crashing feels good, getting out of trouble feels good. A GTA VI that ships with a so-so feel and a stunning prologue mission would be a much weaker long-tail product than a GTA VI that ships with a sublime feel and a merely competent prologue. The composition of the 2022 internal build, as described in mainstream press, implies that Rockstar's leadership understands this and is allocating engineering and animation time accordingly.
It is also worth noting, as inference, that Rockstar has unusual freedom to make this bet. Most studios cannot afford to spend the early-to-mid production years on systems without producing a vertical slice, because they need a vertical slice to keep a publisher or platform-holder on board. Take-Two Interactive's public statements over multiple earnings calls have, by contrast, treated GTA VI as a strategic asset whose schedule is determined by quality rather than by quarterly cadence. The internal latitude that produces a leak full of animation tests rather than a leak full of polished missions is itself a structural fact about the studio's commercial position.
The comparison with Red Dead Redemption 2 is essential because it is the most recent comparable Rockstar project for which we have substantial public reporting on production priorities, and because the studio's public statements about RDR2 explicitly foregrounded simulation feel.
Schreier's 2018 Kotaku reporting on RDR2's development — published shortly before the game's release — characterised a studio culture in which kinesthetic and simulation detail were prioritised at the expense of conventional schedule discipline (Schreier, 2018). The reporting catalogued years of overtime spent on animation, environmental simulation and NPC behaviour, and described the studio's leadership as obsessed with the realism of moment-to-moment interactions. The cost of that prioritisation was well-documented in Schreier's reporting: the human toll, the schedule slippage, and the famous quote from a Rockstar co-founder about working "100-hour weeks", which was later partially walked back (Schreier, 2018).
What is striking, in light of the 2022 leak coverage, is the continuity of priority. RDR2's published reception centred on its physicality: the weight of the horse, the deliberation of weapon-draw animations, the friction of mounting and dismounting, the slowness of looting bodies. Critics were divided on whether that physicality was a feature or a flaw — some praised it as immersion, others called it sludgy — but no one disputed that physicality was the spine of the design. The animation system was not a layer beneath the mission design; it was the foundation the mission design sat on.
The 2022 leak coverage, by inference, suggests GTA VI is built the same way, but more so. Reporters described animation, traversal and gunplay tests in the build because those tests were what was being worked on (Bonifacic, 2022; Fanelli, 2022; Phillips Kennedy, 2022). The continuity from RDR2 is clear: physicality first, mission scripting layered on top once the physicality is locked. The inference is that GTA VI doubles down on the RDR2 thesis — that the feel of being inside the simulation is the product — and treats the narrative campaign as one of several payloads delivered through that simulation, rather than as the product itself with the simulation as a delivery mechanism.
A second-order inference is also defensible from the press record: that Rockstar's leadership appears to have concluded, post-RDR2, that the criticisms of RDR2's physicality (sludginess, friction, slowness in routine actions) were design tuning issues within an approach that fundamentally worked, rather than evidence that the approach itself was flawed. If the 2022 leak coverage is read as evidence of the studio's mid-development priorities, the response to those criticisms has been not "do less physicality" but "do more physicality, tuned better". That is consistent with what the press described — animation tests, traversal experiments, gunplay prototypes — being the dominant content of an internal mid-production build.
If the feel-first inference is correct, several things follow for what GTA VI will be at launch — and they are double-edged.
On the upside. The first ten minutes of unsupervised play should be exceptional. If Rockstar has spent disproportionate engineering and animation time on moment-to-moment kinesthetics, basic verbs — walking, driving, shooting, switching protagonists, entering and exiting cover, climbing — should feel notably better than the AAA baseline. The dual-protagonist switching mechanic in particular, which the press described as visible in the leak (two characters working together to rob a diner; Fanelli, 2022), is the kind of system that requires deep animation-state work to feel seamless rather than gimmicky. The press coverage implies that work has been happening at the right level for a long time.
On the downside. A feel-first development order back-loads narrative production. If missions are written and scripted around verbs that are themselves still being tuned for years, then mission content — dialogue, beat-by-beat scripting, cutscene direction, mission-specific scripting — must be compressed into a later production window. That is consistent with the schedule pressure that has been described in subsequent reporting on GTA VI's release timeline. It implies that the risk in this approach is not that the game feels bad — it almost certainly will not — but that the narrative scaffolding feels less worked-over than RDR2's was, simply because there was less calendar time for it to settle.
A specific inferred risk. Where animation systems are the foundation, mission scripting depends on animation states being stable. If Rockstar tuned animation systems late — and the 2022 leak's emphasis on animation testing implies that tuning was still happening — then any mission scripts written against earlier versions of those systems must be re-validated after each animation pass. That kind of churn is invisible to players but expensive in studio time, and it is a defensible inference, from the leak's composition alone, that a significant portion of GTA VI's remaining post-2022 development calendar would have been absorbed by exactly that re-validation work rather than by net-new content.
An inferred upside on bugs. The same prioritisation reduces a category of risk that has hurt other AAA launches: systemic bugs surfacing late because the underlying systems were never properly stress-tested before being wrapped in mission code. A studio that spends years exercising animation, traversal and gunplay in isolation builds up an enormous amount of edge-case data on its own systems before mission scripting starts pulling on those systems in unusual combinations. The press-described composition of the 2022 leak — systems being exercised, not just demonstrated — implies Rockstar is unlikely to ship the kind of foundational systemic instability that has dogged other open-world launches.
The following confidence ratings apply to the inferences in this report. They are deliberately conservative.
High confidence (drawn directly from press reporting, not inference): The September 2022 leak occurred; it consisted of approximately ninety video files from an internal test build; Rockstar confirmed authenticity; Bloomberg's Jason Schreier characterised the material as "early and unfinished"; the press described animation, traversal, gunplay and dual-protagonist interaction clips as prominent in the trove; working developers across the industry characterised the visible production state as structurally normal for that development stage (Bonifacic, 2022; Fanelli, 2022; Phillips Kennedy, 2022; Schreier in Bonifacic, 2022).
Moderate confidence (inference from press-reported composition): That the ratio of content types in the leaked build — heavy on systems and animation tests, light on polished mission content — reflects Rockstar's actual development priorities in mid-2022 rather than being an artefact of what the hacker happened to exfiltrate. The leak is presumed to be a broadly representative sample of an internal build's contents at that moment, but it is possible that polished mission content existed elsewhere in Rockstar's infrastructure and simply was not in the compromised location.
Moderate confidence: That Rockstar's prioritisation of animation and systems work over vertical-slice polish is a deliberate, leadership-level design philosophy rather than an accident of production order. This is supported by the continuity with Schreier's (2018) reporting on RDR2's priorities, but the public record does not contain on-the-record statements from Rockstar leadership describing this approach in the terms used in this report.
Lower confidence (labelled inference): That feel-first prioritisation constitutes a commercial bet — that Rockstar's leadership specifically believes systemic feel sells more GTA than scripted set-pieces. This is a plausible reading of the evidence but is not directly attested in mainstream reporting. It is an interpretation of revealed preference from the press-reported composition of the build.
Lower confidence: That mission content is back-loaded as a consequence of feel-first prioritisation, and that this carries narrative-polish risk at launch. This is a structural inference about production ordering. It is consistent with the press record but is not directly attested.
Speculative, clearly labelled as such: Any specific claim about animation systems, locomotion graphs, blend-tree architectures, capture pipelines, character controllers, or named internal tools is not made in this report. The argument operates strictly at the level of which categories of work Rockstar was doing, as inferred from mainstream press descriptions of what reporters saw. No leaked code, file paths, identifiers, or internal terminology have been used or referenced.
Bonifacic, I. (2022) Massive 'Grand Theft Auto VI' leak shows off early gameplay footage, Engadget, 18 September. Available at: https://www.engadget.com/rockstar-grand-theft-auto-6-leak-155755703.html (Accessed: 14 May 2026).
Fanelli, J. (2022) Massive GTA 6 Leak Unveils Screenshots And Early Footage, GameSpot, 19 September. Available at: https://www.gamespot.com/articles/massive-gta-6-leak-unveils-screenshots-and-early-footage/1100-6507625/ (Accessed: 14 May 2026).
Phillips Kennedy, V. (2022) Developers share work-in-progress footage in solidarity with Rockstar following GTA 6 leaks, Eurogamer, 21 September. Available at: https://www.eurogamer.net/developers-share-work-in-progress-footage-in-solidarity-with-rockstar-following-gta-6-leaks (Accessed: 14 May 2026).
Schreier, J. (2018) Inside Rockstar Games' Culture Of Crunch, Kotaku, 23 October. (Cited via mainstream press summary; original reporting widely referenced in subsequent industry coverage of Rockstar's development priorities.)
Schreier, J. and D'Anastasio, C. (2022) Grand Theft Auto VI (GTA6) Leak Is a Shock to Rockstar Games, Bloomberg, 19 September. Available at: https://www.bloomberg.com/news/articles/2022-09-19/grand-theft-auto-vi-gta6-leak-is-a-shock-to-rockstar-games (Accessed: 14 May 2026).