CoChat
Sign Up
Cited by 39Ethics and Social Impacts of AIAdversarial Robustness in Machine LearningSafety Systems Engineering in Autonomy

Trustworthy agentic AI systems: a cross-layer review of architectures, threat models, and governance strategies for real-world deployment

Open access
ADABARA, IBRAHIM, Olaniyi Sadiq, Bashir, Nuhu Shuaibu, Aliyu, Ibrahim Danjuma, Yale, Maninti, Venkateswarlu·2025·F1000Research
DOI 10.12688/f1000research.169927.1
Open this paper in CoChat — free accountOpen PDF

Abstract

Agentic Artificial Intelligence systems, characterized by autonomous reasoning, memory augmentation, and adaptive planning, are rapidly reshaping technological landscapes. Unlike traditional AI or large language models, agentic AI integrates decision-making with persistent execution, enabling complex interactions across dynamic environments. However, this evolution introduces novel security risks, governance challenges, and ethical considerations that current frameworks inadequately address. This survey provides a cross-layer review of agentic AI, encompassing architectural paradigms, threat taxonomies, and governance strategies. It consolidates findings from adjacent domains such as cybersecurity, AI safety, multi-agent coordination, and ethics, offering a holistic understanding of vulnerabilities and mitigation approaches. We integrate insights from recent advances in defense architectures and governance innovations, highlighting the limitations of static policies in addressing dynamically evolving threats. Real-world deployments from industrial automation to military and policy applications reveal both successful integrations and notable failures, underscoring the urgency of resilient oversight mechanisms. Furthermore, we identify critical research gaps in benchmarking, memory integrity, adversarial defense, and normative embedding, emphasizing the need for interdisciplinary collaboration to develop adaptive, accountable, and transparent systems. This review serves as a narrative synthesis rather than a systematic literature review, aiming to bridge technical, governance, and ethical perspectives. By integrating cross-disciplinary findings, it lays the foundation for future research on securing, aligning, and governing agentic AI in real-world contexts. Ultimately, this work calls for cooperative innovation to ensure that agentic AI evolves as a trustworthy, accountable, and beneficial technology.

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.12688/f1000research.169927.1

Cites

185 of 188 references verified0 citation errors · 0 not checkable

Statistics

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • The agentic AI governance literature emphasizes cross-layer risks, threat models, memory integrity, adversarial vulnerabilities, lifecycle governance, and the insufficiency of static policies for dynamic systems.

    “The agentic AI governance literature already emphasizes cross-layer risks, threat models, memory integrity, adversarial vulnerabilities, lifecycle governance, and the insufficiency of static policies for dynamic systems [25].”

    From Machine Psychometrics: A Mathematical Psychology of Artificial Intelligence (2026) · cites this paper as [25]

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Cites

Flagged first, then by citations

Research works printed but not linked: 4

  1. Saleh A, Tarkoma S, Donta P: Usercentrix: An agentic memory- augmented ai framework for smart spaces. arxiv.org A Saleh, S Tarkoma, PK Donta, NH Motlagh, S Dustdar, S Pirttikangas, L LovénarXiv Prepr. arXiv2505.00472, 2025•arxiv.org. 2025. Accessed: Aug. 02, 2025. Reference Source
  2. Tallam K: Engineering Risk-Aware, Security-by-Design Frameworks for Assurance of Large-Scale Autonomous AI Models. K TallamarXiv Prepr. arXiv2505.06409, 2025•arxiv.org. 2025. Accessed: Aug. 03, 2025 Reference Source|Reference Source
  3. Reuel A, Soder L,B. B.-F. I, et al.: Position: Technical research and talent is needed for effective AI governance. A Reuel, L Soder, B Bucknall, TA UndheimForty-first Int. Conf. Mach. Learn. 2024•openreview.net. 2024. Accessed: Aug. 05, 2025. Reference Source|Reference Source 199.
  4. Birkstedt T, Minkkinen M, Tandon A, et al.: AI governance: themes, knowledge gaps and future agendas. Internet Res. 2023; 33(7): 133-167. Publisher Full Text 207. · DOI

Cited by

Sorted by citations

Related

Authors
ADABARA, I., Olaniyi Sadiq, B., Nuhu Shuaibu, A., Ibrahim Danjuma, Y., & Maninti, V.
Affiliations
Kampala International University (UG)
Corresponding author
ADABARA, IBRAHIM
ORCID
Venue
F1000Research
Publisher
F1000 Research Ltd
Published
2025-09-11
Volume
14
Pages
905
ISSN
2046-1402 (ISSN-L)
DOI
10.12688/f1000research.169927.1
OpenAlex
W4414133540
Cited by
38
References
185
Citations by year
2026: 34 · 2025: 4
FWCI
39.32
Type
journal article
Language
en
Open access
Licence
Locations
F1000Research (publishedVersion, cc-by)
Funders
None
Topics
  • Social Sciences › Safety Research › Ethics and Social Impacts of AI
  • Physical Sciences › Computer Science › Artificial Intelligence › Adversarial Robustness in Machine Learning
  • Physical Sciences › Engineering › Safety, Risk, Reliability and Quality › Safety Systems Engineering in Autonomy
Keywords
agentic AI, trustworthy AI, AI security, threat model, AI governance, AI safety, multi-agent coordination, AI ethics, adversarial defense, memory integrity, AI alignment, AI benchmarking

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review