
UpSafe
Research code for adding controllable safety experts to LLM layers via two-stage upcycling training, with inference-time safety temperature routing…

Research code for adding controllable safety experts to LLM layers via two-stage upcycling training, with inference-time safety temperature routing…
Code, derived data and results for: Evaluating Baseline Detectors of Hijacked LLM Agents Under Rare Attacks and Distribution Shift

Research artifacts for an MCP-based LLM agent that autonomously solves CTF challenges, including ablation data, analysis scripts, and a six-server…

Code and paper source for a NeurIPS EvoRobust study showing simple bag-of-words and metadata features beat most AI control monitors at detecting…

Research artifact and MCP server that audits security and privacy concerns in third-party Discord bots via large-scale review analysis, LLM labeling,…

Research codebase for privacy evaluation of quantized neural networks: trains target/shadow models under PTQ/QAT and runs membership-inference…

CybORG CAGE 2 environment updated for the paper "Alignment of Autonomous Cyber Defense Agents via Adaptive Policy Shaping"

Benchmark code and interactive results for autonomous network-attack detection by tool-using LLM agents investigating raw PCAPs.

Research code, benchmark and defense for target rebinding attacks that trick LLM safety guards into grading forged decoy content, plus segment-max…

Replication package studying how LLM vulnerability auditors defer to static-analysis evidence, with a tree-sitter C/C++ front end, controlled…

Research code for measuring jailbreak attack success rates (ASR and stepwise ASR) against LLMs, with implementations of attacks such as GCG and…

Jailbreak vulnerability across early-exit depths LLMs. Published at NeurIPS 2026 in the FLMSec workshop

Reference implementation of privacy-preserving face recognition using replicated secret sharing in MP-SPDZ, with quantized GhostFaceNet inference,…

Reproducible Colab pipeline that grounds six AI Security Literacy competency dimensions in public benchmarks and survey microdata, producing figures,…

Diagnostic LLM safety benchmark with 3,489 bilingual prompts, an 84-label taxonomy, and prompt-level rubric judging for explainable, fine-grained…

Automated framework for frontier AI safety evaluation that synthesizes executable, pressure-rich test environments to uncover latent risks in…

Runtime guard skill that detects and blocks malicious skill-induced attacks in skill-augmented agents, with an MCTS loop that evolves the guard…

CyberRAG — RAG 100% local (Ollama) sobre apuntes y libros de ciberseguridad: recuperación híbrida, respuestas con citas, asistente de pentest por…