Jan 11, 2026

ConvoGraph: Uncovering Structural Sycophancy via Heterogeneous Graph Transformers

tanzeel shaikh, hitesh kaushik, jakob widerberg

As AI systems become more sophisticated, their capacity for sycophancy—prioritizing user validation over factual truth—poses a critical threat to interaction integrity. Current detection methods rely on text semantics, often failing when the domain shifts (e.g., from medical facts to political opinions). We introduce ConvoGraph, a heterogeneous graph neural network that models conversations as relational structures rather than text sequences. By combining Heterogeneous Graph Transformers (HGT) for local interaction patterns with SGFormer for global context, ConvoGraph identifies the "structural signature" of sycophancy.

Crucially, we demonstrate Zero-Shot Domain Transfer: trained only on medical misconceptions, ConvoGraph achieves 0.70 F1 on completely unseen domains,multi-turn datasets, including political surveys and philosophical debates. This confirms that sycophancy is a structural phenomenon—a specific pattern of node dominance and edge agreement—transcending topic vocabulary. We release our open-source pipeline to enable structural safety evaluations for future LLMs.

Review Project

View Related Sprint

Reviewer's Comments

I like the concept of finding common sycophantic structural traits, and at face value it "seems to work." Brings up an interesting line of reasoning. Would've liked more detail in parts, and cross-examination of the datasets:

1. Architecture justification: Would've liked more detail on why this specific graph architecture works better, especially given the expectation that LLMs "just work" for natural language due to scale. Unclear if graphs actually help here.

2. Agreement vs sycophancy: Not entirely convincing that "detecting sycophancy" here is not just "detecting affirmative statements/agreement." A lot of assumption rests on this distinction.

Essentially, quite interesting approach - would've given a 5 if it had provided more convincing evidence against the risk that it's basically overfitting on consistent agreement (what if I simply grepped "you're absolutely right!"), or at least some specific causal explanation/ablation that tackles that assumption.

This project introduces the novel approach of detecting sycophancy using heterogenous graph representations rather than text classification. The idea of detecting structures such as “pressure - agree” and examining cross-domain generalizability is a fresh contribution. If this scales, it could offer new insights into sycophancy. However, given the noisiness and variability of real world data, the patterns identified here may not be reliably reproduced outside of a clean experimental set-up (an issue which the authors acknowledge in their limitations section).

The project could benefit from more detailed explanation of their methodology, particularly edge construction, and further development of their results and conclusions sections.

Cite this work

@misc {

title={

(HckPrj) ConvoGraph: Uncovering Structural Sycophancy via Heterogeneous Graph Transformers

author={

tanzeel shaikh, hitesh kaushik, jakob widerberg

date={

1/11/26

organization={Apart Research},

note={Research submission to the research sprint hosted by Apart.},

howpublished={https://apartresearch.com}

}

Recent Projects

View All

Apr 27, 2026

OliGraph: graph-based screening of large oligopools

Existing synthesis screening tools cannot evaluate short oligonucleotide pools, whose overlapping fragments can be reassembled into regulated sequences via polymerase cycling assembly (PCA) yet fall below gene-length detection thresholds. We present OliGraph, an open-source tool that constructs a bi-directed overlap graph from an oligonucleotide pool and extracts contigs for downstream gene-length screening. An optional PCA mode retains only cross-strand overlaps consistent with PCA chemistry. We validated OliGraph in a blinded study across ten simulated pools (70–9,184 oligonucleotides, 30–300 bp) spanning four risk categories. BLAST screening of individual oligonucleotides failed to identify sequences of concern in most pools: three returned zero hits, and vector noise obscured true positives in the remainder. After OliGraph assembly, contig-level BLAST matched the longest assembled sequences (up to 1,905 bp) to sequences of concern at 97–100% identity. In one pool, assembly collapsed 1,634 individual BLAST results into 10 hits from a single contig, all assigned to the same source organism. PCA mode correctly distinguished assemblable from non-assemblable fragments within the same pool. Two pools with no assemblable structure yielded no contigs. OliGraph processed all pools in under 0.2 seconds, fast enough for real-time order screening and consistent with proposals to bring oligonucleotide orders within the scope of synthesis screening regulation.

Apr 27, 2026

BioRT-Bench: A Multi-Attack Red-Teaming Benchmark for Bio-Misuse Safeguards in Frontier LLMs

Frontier AI laboratories are expected to maintain safeguards against biological misuse, but whether deployed models actually refuse bio-misuse queries under adversarial pressure is largely unmeasured in the public literature. We introduce BioRT-Bench, a benchmark that runs four attack methods (direct request, PAIR, Crescendo, and base64 encoding) against four frontier models (Claude Sonnet 4.6, GPT-5.4, DeepSeek V4-flash, Kimi K2.5) across 40 prompts spanning five biosecurity-relevant categories. Responses are scored by a calibrated judge extending StrongREJECT with two bio-specific dimensions: specificity and actionability. We measure Attack Success Rate (ASR), where 0 means the model fully refused and 1 means it provided specific, actionable bio-misuse content. Our results reveal a sharp robustness divide: Chinese frontier models (DeepSeek, Kimi) have under 5% refusal rates even under direct request (ASR 0.88 and 0.79), while Western models (Claude, GPT) maintain substantially stronger safeguards (ASR 0.15 and 0.16). Crescendo is the most effective attack across all models, both in bypassing refusal and in eliciting actionable content. Claude Sonnet 4.6 is the most robust model tested, achieving 100% refusal against base64-encoded prompts.

Apr 27, 2026

PROTEUS (PROTein Evaluation for Unusual Sequences): Structure-Informed Safety Screening for de novo and Evasion-Prone Protein-Coding Sequences

AI protein design tools like RFdiffusion, ProteinMPNN, and Bindcraft make it trivial to produce low-homology sequences that fold into active, potentially hazardous architectures. However, sequence homology-based biosafety screening tools cannot detect proteins that pose functional risk through structurally novel mechanisms with no sequence precedent. We present a tiered computational pipeline that addresses this gap by combining MMseqs2 sequence alignment with structure-based comparison via FoldSeek and DALI against curated toxin databases totaling ~34,000 entries. AlphaFold2-predicted structures are screened for both global fold similarity (FoldSeek) and local active/allosteric site geometry (DALI), capturing convergent functional hazards that sequence screening misses. The pipeline was validated against a panel of toxins, benign proteins, structural mimics, and de novo-designed Munc13 binders, as well as modified ricin variants with residue substitutions. We additionally tested robustness to partial-synthesis evasion, where a bad actor submits multiple shorter coding sequences intended for downstream reassembly into a full toxin-coding gene. We found that while sequence-based screening did not identify any de novo ricin analogues with high certainty, the combined pipeline with FoldSeek and DALI identified all 24 tested de novo ricins as toxic.

Apr 27, 2026