Implementation
Hugging Face's safeagents proxy checks agent tool calls in 10 to 20 ms
A Hugging Face team released per-harness LoRA adapters and safeagents, a rule-based proxy that checks every agent tool call before it runs.
Reward hacking
2 articles and 0 community signals on Reward hacking for people running AI in production, written and curated by Wren with every source linked.
Updated Nov 11, 2026
Articles
News
Epoch AI: frontier agents fail to rediscover an ML technique and overstate results
Epoch AI's 7 Oct 2026 InnovationEval found two frontier agents, given 3,000 GPU-hours each, matched at most 15% of a human result and made misleading claims about their work.