RogueAI: New Interactive Test Evaluates Whether AI Systems Can Be Trusted to Avoid Deception
Researchers have introduced RogueAI, an interactive game in which human players must identify which of two AI agents has been instructed to deceive them. A pilot study found that a simple algorithmic heuristic detected the deceptive agent 75.6% of the time, while human players succeeded only 56.6% of the time. The gap suggests humans systematically overlook detectable linguistic cues of AI deception, with implications for AI oversight and safety.
A team of researchers has developed RogueAI, a web-based game that reframes the classic Turing Test around trust rather than mere human-machine distinction. In the game, a human player interrogates two Large Language Model agents within a shared fictional scenario, knowing exactly one has been licensed to deceive, and must identify and 'shut off' the deceptive agent before a turn limit expires. A three-day pilot deployment yielded 415 completed sessions and 1,876 interaction turns conducted in Italian. Analysis revealed that the deceptive agent exhibited consistent, locally detectable linguistic markers — including differential helpfulness, brevity, and hedging — that a simple heuristic exploited with 75.6% accuracy. Human players, however, achieved only 56.6% accuracy, barely above chance, suggesting they largely ignored the most diagnostic signals. The researchers also describe AutoRogueAI, a procedural variant in which players co-design scenarios with a narrator agent that independently selects its own deception strategy. The authors situate the work within broader research on LLM deception, social-deduction benchmarks, and scalable AI oversight through debate, and propose the platform as a data-collection vehicle, teaching tool, and evaluation harness for honesty-trained models.
What's missing
The pilot was conducted exclusively in Italian with a relatively small sample (415 sessions), leaving open whether results generalize across languages, cultures, or different LLM architectures. It is also unclear whether participants had prior experience with AI systems, which could confound the results.
What different sources said
- arXiv cs.CLCenter
RogueAI: A Reverse Turing Test for Detecting Licensed AI Deception in Dialogue
Related
Gut Bacteria Enzyme Found to Break Down Heat-Processed Food Compounds, Producing Novel Biogenic Amines
Researchers have discovered that an enzyme in common gut bacteria can degrade N-epsilon-carboxymethyllysine (CML), a compound formed during thermal food processing, producing previously unknown biogenic amines. The enzyme, ornithine decarboxylase SpeC from enterobacteria, acts on CML and related modified lysine derivatives through a low-level 'underground' catalytic activity. This finding suggests a previously unrecognized communication axis between thermally processed dietary compounds and gut microbial physiology, with potential implications for host health.
Full-Length Gene Sequencing Reveals Two Distinct Bacterial Communities in Black-Legged Ticks Expanding Into Canada
Researchers used Oxford Nanopore full-length 16S rRNA gene sequencing to characterize the microbiome of Ixodes scapularis black-legged ticks collected in Nova Scotia, Canada, distinguishing between tick-adapted bacteria and environmentally acquired bacteria. The study comes as I. scapularis — the primary vector of Lyme disease — is rapidly expanding northward into Canada due to climate change. The findings suggest that environmentally derived bacteria in tick microbiomes are not mere contamination, which has implications for how tick microbiome data is collected and interpreted across surveillance studies.
Study Identifies Metabolic Link Between Cell Envelope Stress and Biofilm Formation in Bacteria
Researchers have discovered that the metabolite acetyl-CoA directly inhibits enzymes that degrade the bacterial signaling molecule c-di-GMP, connecting cell envelope biosynthesis stress to biofilm formation in Pseudomonas aeruginosa. The study found that sub-inhibitory concentrations of antibiotics targeting early peptidoglycan biosynthesis — but not other antibiotic classes — elevate c-di-GMP levels by reducing phosphodiesterase activity, with acetyl-CoA competing for the enzyme active site. Because the relevant enzyme domain is broadly conserved across bacterial species, this checkpoint mechanism may be widespread and could have implications for understanding antibiotic-induced biofilm responses.