Formalizing authorization, revocation, and supply-chain control for long-running AI agents at Georgia Tech
3
papers here
3
reviewed by Evolve
3
first-author
2
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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 metrics are available yet in the current snapshot: total publication count, venue distribution, and preprint-to-publication conversion are not recorded. This axis would measure the volume and cadence of scholarly output across venues and years.
Impact—
Impact cannot be scored yet: citation counts, h-index, and downstream uptake are absent from the provided bibliometric data. This axis would measure how widely the three 2026 preprints shape follow-up work on AI-agent authorization and security.
Focus0/100
The focus score is 0, reflecting an even three-way split in the field distribution: AI agent security (access control / formal methods), systems security (AI agent supply-chain security), and security and distributed systems (AI agent infrastructure) each hold exactly one third (0.333) of the three-paper portfolio. The spread is broad within a narrow band — all three subfields attack the same object, the AI agent, from authorization, supply-chain, and infrastructure angles.
Collaboration22/100
Collaboration scores 22, consistent with a very small network: the snapshot records only 2 distinct co-authors across three papers, and Chu Wang is the single recurring partner. All three 2026 preprints are first-authored, two also corresponding, indicating an individually driven agenda where formal writing and protocol design are led by Zhu rather than distributed across a team.
Activity—
Activity data is insufficient: per-year publication counts and research velocity are not part of the current record, which covers only 2026 preprints. This axis would measure sustained publishing cadence and engagement across multiple years.
Reproducibility43/100
Reproducibility scores 43, a mean computed across 3 papers. The portfolio pairs formal proofs with executable artifacts — registered tests in the runtime-authorization work and a commit-time reference monitor in ClosureBound — yet Evolve's verdicts note the evidence base is offline and synthetic, and readiness ranges from medium to low across the three papers.
Research summary
Genliang Zhu is a security researcher at Georgia Institute of Technology whose recent work builds formal guarantees for the authorization life cycle of long-running AI agents. Across three 2026 arXiv preprints — all first-authored, two with corresponding role — Zhu addresses three complementary layers of agent security: runtime authorization for resources that agents acquire on a user's behalf, revocation of authority in systems with delegation and asynchronous execution, and supply-chain binding of agent skills to exact versioned dependency closures. The methodological signature is consistent: define the problem formally, state conditional theorems with proofs (eight in the runtime-authorization paper), and pair them with executable artifacts such as registered tests, a root-scoped quiescence protocol, and ClosureBound, a commit-time reference monitor. Evolve's review rates the portfolio's reproducibility at a mean of 43 across three papers, while noting that empirical evidence remains offline and synthetic rather than provider-attached. Co-authorship is narrow — two distinct co-authors, with Chu Wang the recurring partner — and the field distribution splits evenly across AI agent security, systems security, and security and distributed systems, marking an early-stage but tightly focused research program.
Collaboration pattern
The visible network is minimal: 2 distinct co-authors across the three catalogued 2026 papers, with Chu Wang the only recurring name. Zhu is first author on all three and corresponding author on two, so the agenda, proofs, and artifacts appear self-led. This is an early-stage collaboration footprint — joint work seems confined to one close partner rather than a lab-scale or cross-institution team.
Research keywords
Highlights
Eight conditional proofs for runtime authorization of agent-acquired resources
Root-scoped quiescence certifies agent authority revocation
ClosureBound: commit-time binding of skill grants to dependency closures
Timeline
- 2026Preprints ClosureBound, a commit-time monitor binding agent-skill grants to versioned transitive dependency closures.
- 2026First-authored preprint formalizes runtime authorization for resources acquired by AI agents, with eight conditional proofs and registered tests.
- 2026Proposes root-scoped quiescence, a formal protocol for authorization revocation under delegation and asynchronous execution.
Papers in the Evolve catalog
- first authorcorresponding
Authorization Revocation for Long-Running AI Agents: Root-Scoped Quiescence under Delegation and Asynchronous Execution
- first authorcorresponding
Runtime Authorization for Resources Acquired by AI Agents
- first author
Versioned Transitive Dependency-Closure Binding and Operation-Time Effect Governance for Agent Skills: ClosureBound
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.
- Profile is name-resolved from public bibliographic data; identity is anchored to ORCID 0009-0007-1054-5097 and the listed Georgia Tech affiliation.
- Bibliometrics cover only the 3 arXiv preprints (2026) in Evolve's catalog; no peer-reviewed venue, citation, or h-index data is included.
- The output, impact, and activity radar axes are unscored because their underlying metrics were not available in this snapshot.
- Evolve's reviews describe the papers' evidence as offline and synthetic, so reproducibility scores reflect artifact readiness rather than real-world validation.