CSIRO researcher bridging AI safety and ML security: scalable oversight theory and multimodal poison defenses
2
papers here
0
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
2
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-27) 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—
Output (publication volume and authorship roles over time) cannot be scored yet: the OpenAlex record shows only 7 works in the trailing three years with sparse yearly counts (1 in 2024, 6 in 2026) — too thin for a stable rating.
Impact—
Impact (citations, h-index trajectory, and uptake by other groups) is not yet scorable: the record reports an h-index of 2 and an i10-index of 0 with no citation-by-year series, so influence cannot be reliably measured at this stage.
Focus—
Focus (thematic concentration and continuity) is not yet scorable: OpenAlex topic labels scatter across clustering, facility location and emergency management, topic modeling, neural network applications, and complex network analysis, with no dominant theme consolidated.
Collaboration46/100
Scores 46 on collaboration. The OpenAlex window records 9 distinct coauthors against 7 works over the trailing three years — consistent team-based research rather than solo output. On both Evolve-reviewed 2026 arXiv papers he holds a non-leading position (last author on the AI-debate paper, middle author on TraceGuard, neither corresponding), and the recurring partner list includes Haoyang Li, Yaxin Xiao, Linyan Dai, Jiawen Fu, Zi Liang, Qingqing Ye, Haibo Hu, Rayne Holland, and Liming Zhu.
Activity61/100
Scores 61 on activity. The indexed record is concentrated and accelerating: 6 works in 2026 versus 1 in 2024, with 7 works across the trailing three-year window in total. That front-loaded cadence fits an early-career profile whose steady-state output rate is still forming, and it is the stronger of the two dimensions currently scored.
Reproducibility—
Reproducibility (code, artifact release, and experimental control quality) is not scored: the bibliometric source carries no artifact data, and Evolve marks both reviewed papers only medium repro-readiness — single-seed ablations in TraceGuard, no LLM experiments in the debate paper.
Research summary
Jason Xue is an early-career researcher at the Commonwealth Scientific and Industrial Research Organisation (CSIRO) in Australia whose public bibliographic record — resolved via OpenAlex — points to work at the intersection of AI safety and machine learning security. The record is young: an h-index of 2, an i10-index of 0, seven works over the trailing three years, and a yearly distribution that jumps from one indexed publication in 2024 to six in 2026. OpenAlex topic labels (advanced clustering, facility location and emergency management, topic modeling, neural network applications, complex network analysis) sketch a broad applied-machine-learning background that only partially overlaps the two Evolve-reviewed 2026 arXiv papers. The first, 'When Honesty is Not Enough in AI Debate,' where he is last author, develops a formal SIO framework showing that debate wins can coexist with hidden information leakage in transcripts — rigorous theory without LLM experiments. The second, 'TraceGuard,' where he is a middle author, reports a multimodal poison-filtering defense achieving 98.4% recall across 19 attacks, with single-seed ablations flagged as a caution. Collaboration is team-based, with nine distinct coauthors, and the two dimensions currently scored — activity (61) and collaboration (46) — sit in the middle range.
Collaboration pattern
Collaboration is team-based: the OpenAlex window lists 9 distinct coauthors against 7 recent three-year works (collaboration score 46). Recurring partners include Haoyang Li, Yaxin Xiao, Linyan Dai, Jiawen Fu, Zi Liang, Qingqing Ye, Haibo Hu, Rayne Holland, and Liming Zhu. On the two reviewed 2026 arXiv papers he takes neither first nor corresponding authorship — last author on the AI-debate theory paper and middle author on TraceGuard — consistent with a contributor role inside larger groups.
Research keywords
Highlights
Formal case that honest AI debate can still leak
TraceGuard: high-recall multimodal poison filtering
Timeline
- 2024First publication appears in the OpenAlex yearly counts (1 work), marking the start of the indexed record.
- 2026Indexed output jumps to 6 works, the highest yearly count in the record so far.
- 2026Last-author arXiv preprint 'When Honesty is Not Enough in AI Debate' formalizes hidden-information leakage in debate-based scalable oversight.
- 2026Co-authors TraceGuard, a multimodal data-poisoning filtering defense, on arXiv.
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.
- Bibliometric indicators (h-index, i10-index, yearly counts, topics, coauthor count) come from an OpenAlex identity record only; other databases were not consulted.
- The qualitative assessment draws on just two Evolve-reviewed items, both 2026 arXiv preprints; the OpenAlex topic labels differ from these papers' fields, so the record may reflect broader or only partially resolved activity.
- Jason Xue is a comparatively common name; this profile is name-resolved from public bibliographic data and should be confirmed against ORCID 0000-0002-9172-4252.
- Indicators are early-stage: with an h-index of 2 and i10-index of 0, citation-based dimensions will stay unreliable until a longer publication history accrues.