Policy dependency / Stack layer
Holding Back Ready Agent Turns Instead of Releasing Them Eagerly Cuts P95 Workflow Latency up to 3.50x
arXiv 2609.10964
Stack layer / Threat pattern
EvoSafeHarness searches policies and code together to build a per-model safety harness, cutting attack success from 45.6% to 10.0%
arXiv (2609.05903)
Policy dependency / Threat pattern
MOSAIC Picks a GraphRAG Traversal Policy per Query and Beats the Best Fixed Policy by 9.96 Points
arXiv 2609.11065
Stack layer / Threat pattern
MaP-WAM stores robot memory as completed segment records and turns them into plans, rather than replaying full history
arXiv (2609.11561)
Policy dependency / Stack layer
Senate Negotiators Weigh a 'Duty of Care' Law Letting Federal Courts Block Unsafe Model Releases and Preempting State AI Laws
Reuters
Stack layer / Threat pattern
Skill optimization via contextual bandits cut optimization cost 55-58% using only 50 examples per benchmark
arXiv 2609.11682
Stack layer / Threat pattern
Claude Code 2.1.269 Ships a Plugin Eval Runner and a Knob to Raise the Workflow Tool's Concurrent Agent Cap to 256
Anthropic (claude-code CHANGELOG)
Stack layer / Threat pattern
A Malicious Super-App Can Silently Own Every Mini-App Inside It, and Russia's MAX Demonstrates the Full Set
arXiv 2609.11814