Li Zhang

—

Unclaimed · AI-structured profileOpenAlex

Early-stage researcher spanning LLM-agent security and EEG machine learning, first author of a 2026 security SoK

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-13) 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—

No output score is available yet: publication-volume data is insufficient for this profile. This axis would measure annual paper counts and total production volume; with only two reviewed 2026 arXiv papers on file, it is left unscored rather than estimated.

Impact—

Citation and influence data are not yet available for this profile, so impact is not scored. This axis would measure citation counts, citation velocity, and uptake of the two 2026 arXiv preprints, which are too recent for meaningful citation signals.

Focus0/100

The focus score is 0 out of 100, reflecting an almost perfectly split research portfolio: the field distribution assigns 50% to systems security (LLM agent security) and 50% to computational neuroscience / EEG machine learning. These two areas share little methodological overlap — one is a systematization of agent security doctrine, the other a deep-learning sequence model for EEG — so the profile is currently dispersed rather than deeply specialized.

Collaboration41/100

The collaboration score is 41 out of 100, based on 7 distinct coauthors recorded in the bibliometric data. The frequent-coauthor list — Yang Sun, Jie Shi, Weishan Ye, Yue Pan, Gan Huang, and Zhen Liang — suggests a small, stable working circle rather than a large distributed network. Notably, the list also contains a name identical to the researcher's own ('Li Zhang'), a likely name-disambiguation artifact that cautions against over-reading this metric.

Activity—

Activity data is insufficient to score this axis. It would measure recent publication cadence and ongoing research engagement over time; beyond two 2026 arXiv papers, no multi-year output record exists, so no estimate is made.

Reproducibility38/100

Reproducibility scores 38 out of 100, averaged across the 2 papers Evolve assessed. The drivers of this modest mean are visible in the reviews: the EEG tokenization paper carries medium reproducibility readiness but was flagged for missing significance tests and a contradictory Table 2, while the first-authored security systematization is rated low readiness, since a conceptual SoK offers no reusable artifact. It is the weakest evidenced dimension so far.

Research summary

Li Zhang is an early-stage researcher whose reviewed output to date spans two distinct areas: systems security for LLM-based agents, and machine learning for EEG within computational neuroscience. The Evolve catalog holds two 2026 arXiv papers. The first-authored 'When the Agent Becomes the Kernel' systematizes LLM-agent security by mapping it onto operating-system complete-mediation doctrine; Evolve calls it conceptually sharp and well-evidenced while flagging that its central irreducibility claim is argued rather than proven. The co-authored 'Brain-Token Learning' proposes microstate-based tokenization and multi-scale interaction for long-horizon EEG sequence modeling, reporting broad cross-dataset gains that the review tempers with missing significance tests, absent modern baselines, and a Table 2 contradiction. Bibliometric radar scores are modest: collaboration 41/100 over 7 distinct coauthors, reproducibility 38/100 averaged across 2 assessed papers, and focus 0/100 reflecting an even 50/50 split between the two fields. Output, impact, and activity lack sufficient data to score. No institution, country, or ORCID is available, and given how common the name is, this profile is name-resolved from public bibliographic data only.

Collaboration pattern

Collaboration is modest: the collaboration score is 41/100 over 7 distinct coauthors. The recurring circle includes Yang Sun, Jie Shi, Weishan Ye, Yue Pan, Gan Huang, and Zhen Liang, consistent with small-team authorship on both the security SoK (first-authored) and the EEG paper (middle author). The coauthor list also contains a second 'Li Zhang', likely a name-disambiguation artifact, so network metrics should be read with caution.

Research keywords

LLM agent securityPrompt injectionReference monitorComplete mediationTrust boundariesAI-native operating systemsSecurity systematization (SoK)EEG sequence modelingEEG microstatesTokenizationLong-horizon sequence modelingCross-dataset generalization

Highlights

First-authored SoK maps LLM-agent security onto operating-system doctrine
In 'When the Agent Becomes the Kernel' (2026, arXiv), Li Zhang leads a systematization of knowledge that frames LLM-agent security through the classic complete-mediation principle and reference monitors, analyzing trust boundaries on the path to AI-native operating systems. Evolve's verdict: conceptually sharp and well-evidenced, though the central irreducibility claim is argued rather than proven; reproducibility readiness is low.
Microstate-based tokenization for long-horizon EEG modeling
The co-authored 'Brain-Token Learning' (2026, arXiv) proposes microstate-based tokenization with multi-scale interaction for long-horizon EEG sequence modeling and reports broad cross-dataset gains. Evolve tempers this: significance tests and modern baselines are missing, and a Table 2 contradiction weakens the claims; reproducibility readiness is medium.
An even split across two unrelated fields
The bibliometric field distribution assigns exactly 50% to systems security (LLM agent security) and 50% to computational neuroscience / EEG machine learning, yielding a focus score of 0/100. Both catalog papers appeared in 2026, making this dual-track profile very early-stage.

Timeline

2026First-authored systematization 'When the Agent Becomes the Kernel' posted to arXiv, mapping LLM-agent security onto OS complete-mediation doctrine.
2026Co-authored 'Brain-Token Learning' posted to arXiv, introducing microstate-based tokenization and multi-scale interaction for long-horizon EEG sequence modeling.
2026Both catalog papers reviewed by Evolve, with reproducibility readiness rated low (security SoK) and medium (EEG model).

Papers in the Evolve catalog

Frequent co-authors

Yang Sun×1 Jie Shi×1 Weishan Ye×1 Yue Pan×1 Gan Huang×1 Zhen Liang×1 Li Zhang×1

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.

Name resolution: 'Li Zhang' is an extremely common name; no ORCID, institution, or country is available, so this profile is resolved from public bibliographic data and may mix same-name records.
Radar dimensions output, impact, and activity are unscored due to insufficient bibliometric data.
Evidence base is two 2026 arXiv preprints reviewed by Evolve; neither is peer-reviewed and no citation data is included.
The frequent-coauthor list contains a name identical to the researcher's own, indicating co-author disambiguation noise.