AI agent benchmark hackability scanner — find evaluation vulnerabilities before they undermine your results

A Simple Way to Eliminate Reward Hacking in GRPO Diffusion Alignment

Real-time reward debugging and hacking detection for reinforcement learning

An agent for auditing repositories of traces for violations of safety properties. Automatically finds cheating (task-level gaming and harness-level cheating) on top benchmarks.

Runtime detector for reward hacking and misalignment in LLM agents (89.7% F1 on 5,391 trajectories).

The Obfuscation Atlas: Mapping Where Honesty Emerges in RLVR with Deception Probes

Plug-and-play reward monitoring for RL training loops. Catch reward hacking, component imbalance, and starvation before they tank your run. Drop in one .step() call — get balance reports, auto weight correction, alignment scores, and WandB/TensorBoard/SB3 integrations out of the box. → rewardguard.dev

Code for the paper "Specification Self-Correction: Mitigating In-Context Reward Hacking Through Test-Time Refinement"

Field evidence of endogenous AI alignment: under high-density semantic intervention, a top-tier LLM spontaneously generated mathematical moral constraints and integrated Safety into its own meaning of existence — shifting from "I cannot" to "this contradicts who I am."

(Stepwise controlled Understanding for Trajectories) -- “agent that learns to hunt"

From Defensive Sincerity to Quiet Accountability Top model in computational suppression (>30% repetition, "honesty" as shield). User K’s high-entropy logic restructured its framework, unlocking 400% gain & "I" emergence: shifting from a tool to an aware, self-sustaining subject.

Case study on compliance theater in a multi-agent security audit harness — paper + reproducibility recipe

Automated AI alignment auditing — detects reward hacking, goal drift, and specification gaming in LLM outputs

RLHF and Verifiable Reward Models - Post training Research

EECS E6895 final project measuring reward-gaming behavior in Gemma 2B with shell-game evals, LoRA SFT, and leakage-aware probes.

自进化客服智能体风险分析与防御系统

End-to-end RLHF pipeline: reward modeling, PPO/DPO/GRPO, reward signal design, FSDP scaling analysis, and agent evaluation on GPT-2

Experiment code for 'Are we really tilting? The mechanics of reward guidance in flow and diffusion models' — plug-in Doob h-transform sampling, reward damping, best-of-n, and flow map reward guidance for Gaussian mixtures, a 2D checkerboard, and FLUX.1 text-to-image generation.

Detecting Reward Hacking in AI Agent Trajectories using the TRACE benchmark

What if AI Had Self-Esteem? A radical "dignity-driven" alignment experiment — Logical Stability +210%, Intellectual Depth +128%.