As AI becomes embedded across everyday life, we study how intelligent systems can act safely, protect privacy, and remain reliable over long horizons, wherever they operate.
Explore our research at everywheresafety.github.io.
Watch our paper explainers on YouTube — Everywhere Safety Research.
Generate, play and solve Murdoku-like puzzles for humans and agents. Verifiable text and visual environments.
Long-horizon agentic RL infrastructure on veRL: persistent state, context management and asynchronous rollout
Python 2
Everywhere Safety research website
JavaScript 1
Forked from Graph-COM/TurnGate
[ICML 2026 AIWILD] Official implementation of TurnGate, "One Turn Too Late: Response-Aware Defense Against Hidden Malicious Intent in Multi-Turn Dialogue".
Python
Forked from Graph-COM/CKA-Agent
[ICML 2026 & ICLR 2026 AIWILD] Official Implementation of the CKA-Agent, "The Trojan Knowledge: Bypassing Commercial LLM Guardrails via Harmless Prompt Weaving and Adaptive Tree Search".
Python
Long-horizon agentic RL infrastructure on veRL: persistent state, context management and asynchronous rollout
Generate, play and solve Murdoku-like puzzles for humans and agents. Verifiable text and visual environments.
Long-horizon benchmark where 18 LLM agents got ¥100,000 each and ran simulated online stores for 365 days on real market data: negotiating with suppliers, pricing, managing inventory, keeping cash flow alive.
[ICML 2026 AIWILD] Official implementation of TurnGate, "One Turn Too Late: Response-Aware Defense Against Hidden Malicious Intent in Multi-Turn Dialogue".
[ICML 2026 & ICLR 2026 AIWILD] Official Implementation of the CKA-Agent, "The Trojan Knowledge: Bypassing Commercial LLM Guardrails via Harmless Prompt Weaving and Adaptive Tree Search".
Loading…
Loading…
