AIICD
Behavioral detection / Flagship experimentCursor data, learned representations, and motion features for internal-cheat detection.
Open sourceMIT · ExperimentalSentinelAC Archive
SentinelAC was an anti-cheat system built and operated for FiveM. This archive documents the detection systems, experiments, and tooling developed through that work.
The commercial project has concluded. AIICD and the SentinelAC Open Source Edition are now publicly available.
AIICD / Behavioral detectionExperimental · MIT licensed
AI Internal-Cheat Detection
Internal cheat menus can change how a player interacts with the client. AIICD explores those patterns through short windows of cursor movement, using data available to a FiveM resource.
A Temporal CNN learns a representation of the movement sequence. XGBoost combines that representation with engineered motion statistics to produce a score for review.
01 / CollectCursor samplesFiveM client → server
resource/client.lua
02a / LearnTemporal CNNRaw sequence → embedding
02b / DescribeMotion featuresSpeed, jerk, angles, reversals
03 / ClassifyXGBoostEmbedding + features
main.py / score()
04 / ReturnA detection signal
prob_cheat
labelFor server-side review
Collection and features in Lua. Local inference in Python.Architecture notes
The interesting part
Two views of the same movement.
Learned temporal patterns sit alongside explicit motion statistics. The model can use both the sequence itself and features a developer can inspect.
The release
Follow it from sample to score.
The FiveM collection resource, feature engineering, FastAPI service, model artifacts, and documentation are available together. The training dataset is not included.
The next work
Evaluate. Question. Extend.
Inspect the feature extraction, evaluate on representative local data, or contribute sanitized fixtures and evaluation notes. Start with the model card and contribution guide.
A signal, not proof. Cursor behavior is indirect evidence. False positives, changing cheat behavior, and differences between servers remain part of the problem. AIICD is experimental and should not be the sole basis for permanent punishment.
Background
About the original project
SentinelAC began as a commercially operated anti-cheat for FiveM. Its work crossed client-side detection, server-event validation, resource integrity, and the practical business of reviewing suspicious behavior.
The Open Source edition preserves that development. AIICD takes one experiment further into focus: what cursor movement can tell us about internal cheat-menu use, and where that signal stops being reliable.
The work
A small, open research index.
Two repositories. Different parts of the same development story.
SentinelAC
Open Source / FiveM resourceLayered detection, guarded events, resource checks, and a self-hosted API. The broader system the research came from.
Open sourceMIT · Open Source EditionInside the SentinelAC source
Event protectionGuarded event registration and token/hash checks, alongside server-side validation.Read the architecture ↗
Layered detectionsMovement, camera, combat, entity, and resource checks with documented false-positive considerations.Read the detection guide ↗
Self-hosted APIConfiguration, bans, logs, and telemetry served through the Open Source Edition’s local API.Read the API reference ↗
Development notes
What the work taught us.
-
01
Every project has limitations to overcome.
Internal menus can intercept clicks before the resource sees them. AIICD’s client disables click capture; the experiment focuses on movement. A missing input changes what a model can reasonably claim.
AIICD / Design reasoning ↗ -
02
Different setups can affect detections.
A teleport, camera change, or weapon grant can come from an admin tool or a custom game resource. SentinelAC’s detection guide documents these overlaps because a suspicious event needs context.
SentinelAC / False-positive considerations ↗ -
03
A protected event still needs validation.
Wrapping an event is one layer. The receiving resource still needs to check permissions, ownership, distance, state, and cooldowns.
SentinelAC / Event protection ↗
Open source
We Open-Sourced
for a reason.
We’re opening this work so it can be read, questioned, and built on. Preserving the implementation makes the lessons useful to other developers working on game security and behavioral detection.
AIICD and the SentinelAC Open Source Edition are released under the MIT License. These are edited public releases with a defined scope; the original private infrastructure is not part of the release.