anti-cheat-systems

name anti-cheat-systems description Guide for modern game anti-cheat architecture, Windows kernel monitoring, and detection tradeoffs. Use this skill when analyzing EAC, BattlEye, Vanguard, FACEIT AC, kernel callbacks, handle protection, manual-map detection, boot-start drivers, BYOVD, DMA threats, or behavioral telemetry in game security research. Anti-Cheat Systems & Analysis Overview This skill covers layered anti-cheat design across kernel drivers, privileged services, in-game components, and backend telemetry. It is most useful for mapping how modern anti-cheats monitor process handles, image loads, memory integrity, driver trust, virtualization abuse, DMA threats, and suspicious input behavior on Windows. README Coverage Anti Cheat > Guide Anti Cheat > Stress Testing Anti Cheat > Driver Unit Test Framework Anti Cheat > Anti Debugging Anti Cheat > Page Protection Anti Cheat > Binary Packer Anti Cheat > CLR Protection Anti Cheat > Anti Disassembly Anti Cheat > Sample Unpacker Anti Cheat > Dump Fix Anti Cheat > Encrypt Variable Anti Cheat > Lazy Importer Anti Cheat > Anti-Cheat Programming Anti Cheat > Compile Time Anti Cheat > Shellcode Engine & Tricks Anti Cheat > Obfuscation Engine Anti Cheat > Screenshot Anti Cheat > Game Engine Protection:* Anti Cheat > Open Source Anti Cheat System Anti Cheat > Analysis Framework Anti Cheat > Detection:* Anti Cheat > Signature Scanning Anti Cheat > Information System & Forensics Anti Cheat > Dynamic Script Anti Cheat > Kernel Mode Winsock Anti Cheat > Fuzzer Anti Cheat > Windows Ring3 Callback Anti Cheat > Windows Ring0 Callback Anti Cheat > Winows User Dump Analysis Anti Cheat > Winows Kernel Dump Analysis Anti Cheat > Sign Tools Anti Cheat > Backup File / Backup Drivers Anti Cheat > Black Signature Windows Security Features Major Anti-Cheat Systems Easy Anti-Cheat (EAC) Multi-component architecture with service, driver, and game-facing protections Process integrity verification and memory inspection Runtime driver loading with strong client-side enforcement Used by: Fortnite, Apex Legends, Rust BattlEye Kernel driver plus service and game module coordination Handle protection, process monitoring, and memory scanning Strong focus on injected code and runtime tampering visibility Used by: PUBG, Rainbow Six Siege, DayZ Vanguard (Riot Games) Boot-start kernel driver with early visibility into later-loaded drivers Boot-time initialization Driver allowlisting and aggressive system trust checks Used by: Valorant, League of Legends FACEIT AC Kernel-level competitive anti-cheat with strong process and driver monitoring Emphasis on platform integrity and low tolerance for hostile drivers Often discussed alongside Vanguard in kernel anti-cheat research Valve Anti-Cheat (VAC) User-mode detection Signature-based scanning Delayed ban waves Used by: CS2, Dota 2, TF2 Other Systems PunkBuster : Legacy FPS anti-cheat FairFight : Server-side statistical analysis nProtect GameGuard : Korean anti-cheat solution XIGNCODE3 : Mobile game protection ACE (Tencent) : Chinese market protection Detection Mechanisms Detection Decision Methodology Use research-rigor for source verification and empirical validation. Numeric values elsewhere in this skill are examples or research hypotheses unless they are tied to a representative, versioned calibration study for the target game. Define the decision unit: player, engagement, session, account, device, or build; state the game mode, patch, platform, input method, and timeframe. Establish telemetry trust: record whether each field is server-observed, server-derived, client-reported, or reconstructed. Client reports are adversarial inputs; server authority improves trust but does not eliminate clock, replication, schema, or game-logic errors. Keep layers separate: observation -> finding -> attribution -> action. A detector hit is not itself proof of cheating or actor intent. Calibrate locally: derive features, sample floors, and operating thresholds from representative data. Hold out players/sessions and time periods; segment results by relevant populations. Measure deployment risk: report prevalence, FPR, FNR, precision, recall, calibration, uncertainty, and the expected review volume. A score in [0, 1] is not a probability unless calibrated as one. Corroborate correctly: combine causally distinct signals and evaluate their joint errors. Correlated signals, maximum-score aggregation, or a fixed signal count do not guarantee a lower false-positive rate. Review high-impact actions: preserve counterevidence and an appeal path; use human review or independently trusted evidence before punitive action when false positives remain plausible. For invariant findings, first verify that the invariant is guaranteed in the observed state and exclude rollback, retry, reconnect, replication delay, legitimate transitions, administrator/test paths, stale baselines, and game bugs. Describe the result as a state-integrity violation until exploitation and attribution are separately supported. Every evidence package should retain the raw artifact or immutable reference, timestamps and ordering, schema/game/detector versions, feature transforms, threshold/model version, sample counts, provenance, contradictory evidence, limitations, and the exact rule that fired. Memory Detection - Code section hashing and integrity verification - Executable private memory and manual-map detection - Injected module and anomalous image mapping detection - Memory modification and stack provenance monitoring Process Detection - Handle access stripping and protected-process enforcement - Thread start address, APC, and context inspection - Debug register and hidden-thread monitoring - Stack trace and module-correlation analysis Kernel-Level Detection - Driver verification, signature policy, and blocklist checks - Callback registration and object access monitoring - System call, dispatch table, and hook integrity checks - PatchGuard, test-signing, and kernel trust state checks - Kernel pool scanning (Segment Heap aware) for hidden drivers and shellcode Kernel Pool Scanning (Segment Heap Era) Why Segment Heap matters for anti-cheat: Cheat drivers allocate memory in NonPagedPool for shellcode, hook tables, hidden modules. The Segment Heap (19H1+) changed pool internals: headers are HeapKey XOR encoded, allocation paths are split (kLFH, VS, Segment, Large), metadata is isolated. Anti-cheat pool scanners must understand these mechanisms to scan accurately without false positives. Detection targets: 1. BigPool / Large Allocation scanning: - Walk nt!PoolBigPageTable (nt!PoolTrackTable) - Find allocations without corresponding DRIVER_OBJECT or loaded module - Detect manually mapped drivers that allocate large pool chunks - Large allocations have no inline header; metadata is external 2. VS Allocator chunk scanning: - Traverse _SEGMENT_HEAP → VsContext → SubsegmentList - Decode _HEAP_VS_CHUNK_HEADER using HeapKey: real_sizes = encoded_header ^ chunk_address ^ HeapKey - Check decoded chunk for suspicious PoolTag, executable content, or allocation without matching driver - VS chunks carry both _HEAP_VS_CHUNK_HEADER (encoded) and _POOL_HEADER (PoolTag still present) 3. kLFH bucket scanning: - _SEGMENT_HEAP → LfhContext → Buckets[] → AffinitySlots → Subsegments - kLFH randomizes block placement (harder to predict adjacency) - FreeHint encoded with LfhKey - Allocation pattern anomalies in specific size buckets can indicate pool grooming by cheat drivers 4. Suspicious PoolTag detection: - Cheat drivers use custom or rare tags; maintain blacklist - Cross-reference tags against known-good tag database (pooltag.txt) - Tags present in pool but absent from any loaded module = suspicious 5. Executable memory in NonPagedPool: - Find chunks with X permission but no corresponding module - Scan decoded chunk content for known cheat signatures, ROP gadgets, specific syscall stub patterns 6. Segment Heap integrity checks: - Validate the build-specific `_SEGMENT_HEAP` signature/layout using symbols and runtime checks (0xDDEEDDEE is observed on relevant layouts) - Verify VS chunk header encoding consistency - Detect tampered heap metadata (indicates heap exploitation attempt) Required knowledge for scanner: - nt!RtlpHpHeapGlobals (HeapKey, LfhKey) — obtained via pattern scan - nt!ExpPoolQuotaCookie — for ProcessBilled decoding - Per-pool-type _SEGMENT_HEAP instance addresses (nt!PoolVector) - Allocation path determination (size → kLFH/VS/Segment/Large) Anti-cheat KDP integration: - Store detection rule tables in Secure Pool (ExAllocatePool3 + KDP) - Correctly configured KDP can protect selected pages from ordinary VTL0 writes, including kernel R/W primitives, while the hypervisor and policy path remain trustworthy Behavioral Analysis - Raw input timing and pattern analysis - Movement and aim anomaly detection - Statistical improbability and ML-assisted scoring - Telemetry collection and server-side review - AI visual aimbot detection (input pattern + gameplay behavior) AI Visual Aimbot Detection AI visual cheats (screen capture + computer vision + hardware input) can be among the harder classes to detect because some designs avoid game-memory access, code injection, and a cheat driver on the gaming PC. Detection then leans more heavily on trusted behavioral telemetry and contextual signals. Input Pattern Analysis: - Mouse movement micro-signature: a particular automation pipeline may retain acceleration or correction patterns distinguishable from a matched human baseline; this must be demonstrated rather than assumed - Engagement timing: a given automation pipeline may produce a narrower latency distribution than a matched human baseline, but capture, inference, transport, smoothing, frame rate, and input hardware make absolute latency ranges setup-specific - Quantization: a specific coordinate-to-HID conversion may leave repeated rounding patterns, but integer deltas also occur in legitimate input - Correction patterns: some smoothing configurations produce repeated overshoot-and-settle shapes; compare them with matched legitimate behavior - Target switching: an explicit automated scoring objective may produce more consistent ordering than a matched baseline, but implementations vary Gameplay Behavioral Signals: - Anomalous K/D ratio combined with other statistical outliers - "Snap" engagement pattern: rapid crosshair movement to target followed by immediate fire, repeated consistently - FOV-boundary effect: some configured systems produce a sharper engagement cutoff near a chosen radius; estimate it statistically and test alternatives - Consistent headshot angle distribution that doesn't match the player's ranked skill bracket - Engagement rate: compare visible-target engagement with a matched population; high or stable rates are contextual signals, not class rules Environmental Detection: - OBS Game Capture may load a graphics-capture hook into the game on supported paths; this is legitimate capture evidence, not cheat attribution - Window/Display Capture backends vary across Windows Graphics Capture, BitBlt, Desktop Duplication, OBS version, and source settings - Frame-transfer detection should account for shared GPU resources, reusable staging resources, readback, synchronization, and legitimate capture tools - Known hardware input device USB VID/PID signatures (KMBox, certain Arduino/Teensy boards) - USB device enumeration anomalies: input device appearing/changing mid-session - Logitech driver version detection: known exploitable G HUB versions - A known input-filter driver is a contextual signal; legitimate use and actual behavior must be established before assigning risk Server-Side Statistical Analysis: - Aim trajectory reconstruction from server-received input deltas - Compare aim distribution against player population at same rank - Detect systematic per-frame aim correction vectors that deviate from matched legitimate distributions - Cross-session pattern analysis: test whether unusually stable metrics remain discriminative after controlling for skill, hardware, and play style - Replay-based ML classifiers trained on confirmed AI aimbot cases Anti-AI Countermeasures (Game Design): - Evaluate ordinary gameplay effects, visual variety, and UI composition for model robustness without degrading accessibility or legitimate play - Server-side aim validation: reject physically impossible aim transitions - Treat model-targeted visual changes as experiments; adaptive models can retrain, and game-design costs may outweigh temporary detection gains Server-Side Replay Analysis for AI Aimbot Detection Server-side detection can analyze gameplay and input telemetry without relying on local process-scanning hits. It is a strong complementary layer against zero-memory AI cheats when telemetry provenance and integrity are trustworthy; client-uploaded fields remain untrusted until validated. Input Telemetry Collection: - Record raw mouse delta (dx, dy) per tick at server tick rate - Record timestamps at the highest reliable precision supported by the input, engine, transport, and clock-synchronization pipeline - Record crosshair angle / view angle per tick - Record fire events with corresponding view angle at fire time - Record damage events with hit location (head/body/limb) - Collect per-session: total engagement count, hit count, headshot count, K/D, average engagement distance Replay-Based Trajectory Reconstruction: - Reconstruct full crosshair trajectory from recorded input deltas - Overlay trajectory onto 3D game state (player positions, obstacles) - Identify "engagement windows": trajectory segments where crosshair moves toward and locks onto a target - Measure per-engagement: time-to-target, overshoot magnitude, correction count, final hold time before fire Statistical Features for AI Detection: (extracted from reconstructed trajectories) Temporal features: - Reaction time distribution: time from target visibility to first crosshair movement toward target → Compare automation and human distributions only within a matched, versioned setup; target-visibility definition, tick rate, latency, skill, and input method materially change the result - Time-to-lock distribution: time from engagement start to crosshair on target → AI: consistent, speed-limited by smoothing algorithm → Human: highly variable, depends on initial angular distance Spatial features: - Trajectory curvature: some smoothing algorithms produce repeated parametric shapes, but both automation and human trajectories vary by configuration, device, sensitivity, and task - Overshoot-correction ratio: compare distributions within matched conditions; neither automation nor human behavior has a universal shape - End-point precision: test for repeated offsets or concentration relative to a skill- and context-matched baseline - Angular velocity profile: treat smoothness and acceleration as measured features, not class-defining rules Engagement pattern features: - Target selection consistency: automation configured with an explicit scoring objective may select targets more consistently than a matched human baseline; implementations need not use closest-to-crosshair or confidence ordering - FOV boundary effect: AI shows sharp engagement cutoff at configured pixel radius; humans have gradual falloff - Engagement rate: percentage of visible targets engaged; AI engages more consistently than humans who miss, ignore, or react slowly to peripheral targets - Multi-target switch pattern: AI switches with machine-like regularity; humans show grouping and hesitation ML Classifier for AI Aimbot Detection Feature engineering and model architecture for detecting AI-generated mouse input at scale. Feature Vector (per engagement window): f1: reaction_time_ms f2: time_to_lock_ms f3: initial_angular_distance_deg f4: trajectory_curvature_mean f5: trajectory_curvature_std f6: overshoot_magnitude_px f7: correction_count f8: final_hold_time_ms f9: angular_velocity_max_deg_per_sec f10: angular_velocity_std f11: micro_correction_frequency (small deltas < 2px per tick) f12: trajectory_straightness_ratio (distance / path_length) f13: dx_dy_correlation (Pearson correlation of delta components) f14: delta_magnitude_entropy (Shannon entropy of |delta| sequence) f15: fire_timing_relative_to_lock_ms Session-level aggregate features: s1: headshot_ratio s2: hit_ratio s3: reaction_time_cv (coefficient of variation across engagements) s4: engagement_rate (targets engaged / targets visible) s5: k/d_ratio s6: fov_engagement_boundary_sharpness s7: target_selection_optimality_score s8: trajectory_curvature_consistency (inter-engagement variance) Model architecture options: - Gradient Boosted Trees (XGBoost/LightGBM): Strong candidate for tabular feature vectors, with fast inference and useful diagnostics; validate explanations and deployment fit - Random Forest: useful baseline; overfitting depends on data and tuning - 1D-CNN / LSTM on raw delta sequences: Operates on raw (dx, dy, dt) sequences instead of engineered features. Can capture patterns human engineers might miss. Higher compute cost; suitable for batch/offline analysis. - Ensemble: combine tree-based features and sequence models only when held-out evaluation shows a worthwhile gain after calibration and complexity costs Training data: - Positive samples: confirmed AI aimbot users (manual review, honeypot, or controlled testing with known cheat software) - Negative samples: legitimate high-skill players (important: include top-percentile players to avoid false-positives on skilled play) - Hard negatives: players with aim-assist controllers (console), players using legitimate accessibility tools Evaluation metrics: - Choose an operating point from prevalence, error costs, enforcement policy, and review capacity; there is no universal acceptable FPR or TPR - Report FPR, FNR, precision, recall, calibration and confidence intervals on representative held-out data, including population slices - Session-level aggregation can reduce transient noise, but only after validating cross-session dependence, drift, and its effect on both FP and FN Deployment pipeline: Client → input telemetry upload (per tick) → server telemetry DB → batch feature extraction (per engagement window) → ML inference (per session) → risk score aggregation (per player, across sessions) → threshold → manual review queue or automated action Adversarial robustness: - Cheat developers tune smoothing parameters to evade specific features - Defense: retrain model periodically on newly confirmed samples - Use feature combinations rather than single-feature thresholds - Aggregate across sessions only after validating dependence, drift, and attacker adaptation - Evaluate both raw-sequence and engineered-feature models adversarially; neither architecture is inherently harder to evade Hardware Input Device Detection Detecting KMBox and similar hardware input injectors at the platform/driver level. USB Enumeration Signals: - Known VID/PID combinations for KMBox, Arduino Leonardo (2341:8036), Teensy (16C0:0486), generic CH340/CP2102 serial adapters - USB device appearing/disappearing during game session - Multiple HID mouse devices where only one physical mouse is expected - USB device with HID mouse capability but no manufacturer string or generic "Arduino LLC" / "Teensyduino" manufacturer USB HID Report Analysis: - Hardware input devices generate genuine HID reports, but: - Report rate: a simplistic injector may expose programmed periodicity, but real devices and sophisticated injectors can both show jitter - Report timing: some automation pipelines create burst patterns; human and legitimate software-assisted input can also be bursty - Delta distribution: compare against matched devices, polling rates, sensitivity, and movement tasks before drawing conclusions Network Traffic Indicators (KMBox Net): - KMBox Net uses UDP communication on the local network - Packet pattern: consistent-size UDP packets at high frequency from a secondary device to the KMBox's IP - If cheat PC is on the same network: detectable via network monitoring (firewall/router logs) Driver-Level Detection: - interception.sys: known driver signature, detectable via module enumeration and PiDDBCacheTable - Logitech G HUB DLL injection: detect unexpected DLL loads into GHUB process, or specific exploitable GHUB versions via file version checking Limitations: - Protocol-conformant hardware injection may be indistinguishable from a normal mouse from an individual HID report alone; descriptors, timing, provenance, and gameplay behavior can still provide imperfect signals - Device signatures can identify known implementations but are not durable attribution; statistical input analysis also requires calibration - Dual-machine capture can avoid a cheat process on the gaming PC, but still leaves ordinary capture/input-device effects and may leave network or device telemetry depending on the design Anti-Cheat Architecture User-Mode Components Process scanner Module verifier Overlay detector Screenshot capture Kernel-Mode Components Driver loader Memory protection System callback registration Hypervisor and driver trust detection VAD and executable memory inspection Hypervisor-Level Components - EPT-based memory access monitoring - Callback list write protection via EPT hooks - ETW structure integrity enforcement - AC driver code page protection (prevent patching) - VMCALL interface for policy configuration from kernel driver - VM exit handlers for EPT violations on protected regions Server-Side Components Statistical analysis Replay verification Report processing Ban management Research Techniques Static Analysis Dump and analyze AC drivers Reverse engineer detection routines Identify signature patterns Map callback registrations and trust boundaries Dynamic Analysis Monitor system calls Track driver communications Inspect memory layout and module provenance Debug with kernel or hypervisor tools Bypass Categories Memory Access Physical memory read/write DMA-based access Hypervisor memory virtualization Driver-based access DMA Cheat Detection Methodology PCIe-Layer Detection Pipeline Detection targets inconsistencies between what a device claims to be and how it actually behaves. Each technique targets a specific firmware emulation tier or class of gap. Configuration Integrity: - VID/DID/SVID/SDID against known-real-silicon allowlist - Capability-chain walk: DWord-aligned Next pointers, no overlaps, no cycles - Signature-residue scanning: Xilinx 7-series default byte patterns, Device Capabilities field bits, reserved-field defaults - Capability presence consistency: donor model must expose expected caps - BAR mask verification: write 0xFFFFFFFF, compare size mask vs donor BAR Memory Probing: - Send Memory Read TLPs to BAR ranges, validate responses by donor class - NIC BAR0: register layout (ring descriptors, interrupt mask, link status) - NVMe BAR0: controller registers (CAP, VS, CC, CSTS, AQA, ASQ/ACQ) - XHCI BAR0: capability registers (CAPLENGTH, HCSPARAMS, HCCPARAMS) - stock pcileech zerowrite4k returns all-zeros; loopaddr echoes address R/W Consistency: - Toggle writable bits (Command BME, Device Control MPS/MRRS, MSI Enable) - Walk W1C bits (Status, AER Status): write 1s, confirm clear semantics - Walk reserved bits: write 1s, confirm read-back as 0 - Per-register writable masks must match donor model exactly Link-State Validation: - Compare negotiated Width/Speed with slot topology, platform policy, signal integrity, and the claimed device; capable devices can legitimately train down - DLL Active, Slot Clock Config consistency - ASPM behavioral validation only when ASPM is enabled and the workload, observation window, firmware, and platform policy should exercise it AER Baselining: - Compare correctable-error rates against per-silicon baseline - Implausibly clean logs (zero correctables when donor normally produces Bad TLP / Replay Timer Timeout) = anomalous - UR/CA response distribution to probes of unimplemented offsets Completion Latency Fingerprinting Completion-latency distributions can reflect memory, arbitration, buffering, power state, link, driver, and workload behavior. A simplistic BRAM-backed emulator may show unusually low variance, but neither distribution shape is universal. Statistical methods: - Kolmogorov–Smirnov test: compare empirical CDFs vs reference - Hill estimator: tail index (real silicon has non-trivial tail) - Anderson-Darling test: sensitive to tail differences Added random jitter is not equivalent to reproducing a donor distribution. Validate mean, variance, tails, modes, autocorrelation, and condition changes against matched hardware.