AgentLAB — First Long-Horizon Agent Attack Benchmark

Same source domain / Shared topic

ToolHazard synthesizes adversarial agent environments instead of hand-building them.

Reported by the same outlet (arxiv.org); overlapping topics (adversarial, agent, attack, benchmark, environment).

Same source domain / Shared topic / Tension

A Hidden HTML Comment in a Pull Request Owns Your Review Agent, and Microsoft Hasn't Patched It

Reported by the same outlet (arxiv.org); overlapping topics (against, agent, attack, injection); pushes against this story (against).

Multi-turn attacks crack AI agents 5-14% of the time

Reported by the same outlet (arxiv.org); overlapping topics (against, agent, attack, benchmark); pushes against this story (against).

LLM Web Agents Fail Dark Patterns 41–72% of the Time

Reported by the same outlet (arxiv.org); overlapping topics (against, agent, attack, fail); pushes against this story (against).

benchflow-ai/skillsbench — Agent Skills Benchmark (396 stars)

Reported by the same outlet (arxiv.org); overlapping topics (against, agent, benchmark, finding); pushes against this story (against).

Best models hit 70.4% on single-hop API calls and 2.4% on knowing a question is unanswerable.

Reported by the same outlet (arxiv.org); overlapping topics (against, agent, benchmark); pushes against this story (against).

Simulator collapse: RL against one LLM user-simulator doesn't transfer.

Reported by the same outlet (arxiv.org); overlapping topics (against, fail, finding); pushes against this story (against).

Dark Souls Learning Environment: five methods, 22 bosses, every RL agent scores essentially zero.

Reported by the same outlet (arxiv.org); overlapping topics (against, agent, benchmark); pushes against this story (against).