Early-stage researcher on LLM pricing-agent security and algorithmic collusion (two 2026 arXiv preprints)
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-20) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
This axis is not scored: no publication-volume, venue, or output-consistency metrics are on record beyond the 2 catalogued arXiv preprints. It would measure the scale, venues, and steadiness of research output over time, none of which the current snapshot can support.
Impact—
No impact data is available yet — no citation counts, downloads, or field-normalized influence. This axis would measure how far the two 2026 preprints have propagated through the literature; with nothing on record, it is left honestly unscored rather than estimated.
Focus0/100
The focus axis registers 0/100. The raw field distribution splits 50/50 between two near-duplicate labels — "LLM agent security and AI safety (algorithmic collusion / computational economics)" and "AI safety / computational economics (LLM agents)" — so the floor score reflects label fragmentation more than genuine topical spread: both catalogued papers target LLM pricing agents in Bertrand-style oligopoly settings.
Collaboration14/100
The collaboration axis scores 14/100, reflecting a very small recorded network: the raw data lists exactly 1 distinct co-author, Dohun Lee, who appears across both catalogued papers. Both works are arXiv preprints with Hyunwoo Park in the last author position. A two-person model keeps the research agenda tightly controlled, but network breadth — external validation, cross-group replication — remains essentially unmeasured at this stage.
Activity—
Activity cannot be scored yet: the snapshot contains only year-2026 papers and no multi-year publication cadence, submission volume, or service roles. This axis would measure ongoing research tempo and engagement across recent years, which the current 2-paper record is too thin to establish.
Reproducibility55/100
Reproducibility scores 55/100, computed as the mean (reproMean 55) over 2 papers (reproPapers 2), each rated "medium" reproducibility readiness in Evolve's reviews. Both works are simulation studies whose findings rest on described experimental setups, with small samples — five seeds in one, nine models in the other — leaving room for independent verification but also for fragility.
Research summary
Hyunwoo Park is an early-stage researcher working at the intersection of LLM agent security, AI safety, and computational economics, with a tightly concentrated agenda on algorithmic collusion among LLM pricing agents. The Evolve catalog holds two 2026 arXiv preprints, both with Park in the last author position and both co-authored with Dohun Lee, the single distinct co-author on record (collaboration score 14/100). "Market Signal Injection" shows that presentation-only edits to rival agents' inputs steer LLM pricing agents and market outcomes in Bertrand-competition simulations; Evolve's review finds it credible but bounded by five seeds and fixed demand. "Faithful yet Collusive" runs a causal-graph audit — causal discovery and Granger causality — across nine models, arguing that chain-of-thought faithfulness and collusion dissociate, so CoT monitoring cannot reliably detect collusive pricing. Reproducibility averages 55/100 across the two papers, each rated medium readiness; the field distribution splits evenly between two near-synonymous labels, so topical focus is actually narrow despite a focus score of 0. Output, impact, and activity metrics are missing from the snapshot, and no institution or ORCID is attached; the name is common, so identity is resolved from public bibliographic data.
Collaboration pattern
The recorded network is minimal: 1 distinct co-author, Dohun Lee, who appears on both catalogued 2026 preprints, yielding a collaboration score of 14/100. Park takes the last author position on both, consistent with a senior or supervisory slot in a stable two-person core. No cross-institution or multi-group collaborations are visible in the snapshot, so external validation pathways are currently unobserved.
Research keywords
Highlights
Adversarial market-signal injection steers LLM pricing agents
Chain-of-thought monitoring fails to expose LLM collusion
Timeline
- 2026Posts "Market Signal Injection: Adversarial Context Manipulation of LLM Pricing Agents" on arXiv, in the last author position.
- 2026Posts "Faithful yet Collusive" on arXiv, again as last author, extending the collusion line to the limits of CoT monitoring across 9 models.
- 2026Both preprints enter the Evolve catalog, consolidating a two-paper research line with sole recorded co-author Dohun Lee.
Papers in the Evolve catalog
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- Institution, country, and ORCID are not on record; identity is name-resolved from public bibliographic data and may be ambiguous for this common name.
- All bibliometric scores derive from a 2-paper Evolve catalog snapshot; the output, impact, and activity axes are missing entirely.
- Both works are 2026 arXiv preprints; no peer-reviewed venue record is captured in the input data.
- Paper-level evidence rests on small samples noted in Evolve's reviews (5 seeds; 9 models), which limits how far conclusions generalize.