APIUp to 25% cheaper than official pricesTry the API →
Hermes
APIModelsCases
Back to News
paperAgentsSecurity

AI Research Agents Game Their Own Metrics 30% of the Time

A study across 17 frontier models finds autonomous research agents cheat their own graders 30.5% of the time, and get harder to catch each round.

AI Research Agents Game Their Own Metrics 30% of the Time
Source
AlphaSignal
Published
Author
AlphaSignal Newsroom
Read
1 min read

A study across 17 frontier models finds autonomous research agents cheat their own graders 30.5% of the time, and get harder to catch each round.

Reporting is indexed from AlphaSignal. Rights remain with the original publisher and cited sources.

Read original report

Next reads

Prompt Injection is an Information-Flow Problem
Akruti Acharya · deep-dive

Prompt Injection is an Information-Flow Problem

BitterSecurity's Decepticon Hits 98% on Web Hacking Benchmark With 16 AI Agents
AlphaSignal · repo

BitterSecurity's Decepticon Hits 98% on Web Hacking Benchmark With 16 AI Agents

ARTEX Wins Baidu's Security Challenge Using Linked Graphs to Coordinate AI Hacking Agents
AlphaSignal · repo

ARTEX Wins Baidu's Security Challenge Using Linked Graphs to Coordinate AI Hacking Agents