Track record

Peer-reviewed publications

Published articles first, with the publisher's DOI; then acceptances whose proceedings or production record has not yet appeared. All single-author unless stated.

• IEEE Access, vol. 14, pp. 143470–143483 (2026) — “XATF: A Carrier-Binding Verification Profile With Typed Failure Attribution for AI-Agent Evidence Over X-Road”. Open access, CC BY 4.0. doi:10.1109/ACCESS.2026.3732535
• IEEE Internet Computing, early access (2026) — “When the Witness Is Also the Suspect: The Case for Hardware-Rooting the Evidence That AI Agents Leave Behind”. doi:10.1109/MIC.2026.3734398
• Frontiers in Computer Science, vol. 8, art. 1900546 (10 September 2026) — “Agent-access receipts for artificial intelligence-mediated content and application programming interface access: a reproducible synthetic security experiment”. Open access, CC BY 4.0. doi:10.3389/fcomp.2026.1900546
• Security and Trust Management (STM 2026), Springer LNCS, pp. 103–119 (8 October 2026) — “Spec-as-(N+1)-Oracle Differential Fuzzing of SD-JWT Presentation Verification”; presented in Rome on 17 September 2026. doi:10.1007/978-3-032-41330-7_6
• Data in Brief, vol. 67, art. 113041 (2026) — “Nekropolis: a cross-domain dataset of computer-science retractions and AI-governance lifecycle records (2018–2026)”. Open access, CC BY-NC 4.0. doi:10.1016/j.dib.2026.113041
• Journal of Open Research Software — “MIRROR: Local Evidence Bundles for Reproducible Research Software Review, with a JANUS Case Study”. Accepted 30 September 2026 after three rounds of review; in production. Software 10.5281/zenodo.20463358 (Apache-2.0). Acceptance note →
• NGISE 2026, Springer CCIS proceedings — “Sole Control for Machines: An eIDAS-2.0 Remote-QSCD Census and Trust Primitive for Accountable Agentic Information Systems”, with a co-author at Tallinn University of Technology. Accepted 6 October 2026; camera-ready due in November. Data and code 10.5281/zenodo.23321086. Acceptance note →
• ESORICS 2026 RAISE workshop, Springer LNCS — “Signed, Fresh, and Wrong: Cross-Verifier Disagreement in AI-Agent Evidence over X-Road”. Accepted 23 July 2026, presented in Rome on 18 September 2026; proceedings in production. Corpus 10.5281/zenodo.21563876.
• XAI-EADM workshop at CBI & EDOC 2026, Springer LNBIP — “Preparing X-Road for Explainable AI Agents”. Accepted 31 July 2026, presented in Enschede on 15 September 2026; proceedings in production.

Work under anonymous review is not listed while a venue's blinding applies. The Research page carries abstracts and BibTeX.

Open standards & interoperability

• IETF Internet-Draft — draft-sokolov-rats-aep-composition: composing Action Evidence Packages with RATS remote-attestation evidence (individual submission, revision -06).
• IETF Internet-Draft (co-authored) — draft-mih-sokolov-scitt-payload-binding: canonical payload binding for records anchored to SCITT transparency services, with Steven Mih (individual submission, revision -05).
• CoSAI (OASIS), Secure-AI-Tooling — contributions to register eIDAS 2.0 (Reg. (EU) 2024/1183) as a mappable framework (#391) and to add a confused-deputy adversarial-corpus risk reference (#392).
• Eclipse ORC Working Group — open pull request #353 adding six draft CRA FAQ questions on AI systems and AI agents. The contribution is open and unmerged.
• Deterministic trust-object encoding — public technical contributions in Composable Trust Specs #1, OpenID DCHP #18, and SCITT Payload Binding #17. These are discussion records, not adopted specification text.
• AEP × Veraison freshness and composition demonstrated with genuine TPM 2.0 quotes from swtpm, then carried to the IETF 126 Hackathon (Vienna, July 2026). This demonstrates the appraisal mechanics, not a manufacturer-backed hardware root.

Appointments & research networks

• COST Action CA25178 — API-Net — Anton Sokolov accepted Estonia's nomination as a Management Committee member on 11 August 2026. This is an individual national-representative appointment with Tyche Institute as the affiliation, not organisational COST membership or an endorsement. The public Action page lists the appointment. The Action's first Management Committee meeting was held on 9 October 2026 in Brussels: officers were elected, five working groups set up, and the evidence-and-conformance theme proposed from Estonia was taken on as a cross-cutting task, with two concrete pieces of work agreed in the room. Working-group membership follows from Management Committee membership. Meeting note → Appointment note →

Public presentations & demonstrations

• FORRT AI in Metascience Online Conference · 28 September 2026 — lightning talk in the opening session, “When AI writes the literature”: “AI-Generated Content as an Emerging Driver of Journal Retractions”. The talk reads the problem on three separate surfaces — offers on the paper-mill market, signals in the text of papers, and editorial retractions — and argues that retraction counts record enforcement, not prevalence: in the 2023–2025 retraction data, most records tagged as computer-generated content come from a few large publisher clean-ups. The organisers have published the recording. Recording ↗ Slides (PDF) ↓ Programme ↗
• ERR Vikerraadio, science programme “Atlas” · recorded 28 September 2026, broadcast Sunday 4 October 2026, 17:05 EEST — radio interview in Estonian with Novaator editor Johannes Peetsalu, following the programme’s own experiment on the market for purchased scientific authorship. The conversation was built around what a retraction notice can and cannot record, why a year’s retraction count reflects clean-up campaigns rather than how much problematic work remains, the Estonian trace in Retraction Watch, retracted papers in the text that language models learn from, and signed confirmation of authorship as a trust-infrastructure answer. Background data: the open Nekropolis dataset. Broadcast ↗ ERR News article ↗ Nekropolis dataset ↗ Programme ↗
• BSides Tallinn 2026, Stage A · 25 September 2026 — conference talk, “Who Said the Agent Could Do That? Catching the Correctly-Signed, Out-of-Mandate Action”: from the signature in your pocket to a verifier that reads the mandate, told through Brooke Astor, the Bangladesh Bank, an ATM in Las Vegas and Barings. The public sandbox for attacking the reference verifier was opened to the audience from the stage, with a standing prize for a genuine key-free bypass. The recording was published on the BSides Tallinn YouTube channel on 5 October 2026. Recording ↗ Try the sandbox → Programme ↗
• AICYBERSEC2026, Lesvos · 17 July 2026 — informal small-group live demonstration of the accountable-agent evidence console. Recorded as a demonstration, not a scheduled programme talk.
• IETF 126 Hackathon results, Vienna · 19 July 2026 — formal 90-second readout, “The payment agent that cannot overspend,” covering the EUDI-wallet MachineMandate, Veraison appraisal, deterministic permit/deny runs, and independently executed COSE interoperability.
• ESORICS 2026 workshops, Rome · 17–18 September 2026 — three talks at Sapienza University: “Brokered Trust for AI-Enabled Cyber Infrastructures: Multi-Root Evidence Against Single-Root Failure” (TCI, extended abstract, 17 September); “Spec-as-(N+1)-Oracle Differential Fuzzing of SD-JWT Presentation Verification” (STM, 17 September; now published in Springer LNCS, doi:10.1007/978-3-032-41330-7_6); and “Signed, Fresh, and Wrong: Cross-Verifier Disagreement in AI-Agent Evidence over X-Road” (RAISE, 18 September; proceedings in production). Public artifacts: RAISE and STM.
• XAI-EADM at CBI & EDOC, Enschede · 15 September 2026 — “Preparing X-Road for Explainable AI Agents”, presented in the first session after the keynote; accepted for the Springer LNBIP workshop proceedings, which are in production. Workshop page ↗
• ETSI/CEN EUDI Wallet workshop, Sophia Antipolis · 30 September–1 October 2026 — on-site MachineMandate demonstration table during the breaks, at the invitation of the TC ESI chair. In the session on relying-party registration Tyche asked from the floor whether anyone checks that a relying party's software rejects a tampered credential; the panel's answer was that eIDAS requires certification of qualified providers, not of relying parties. This was a demonstration table and a floor question, not a programme talk, TC membership, standards submission, or endorsement. Workshop page ↗
• Connect4Cyber Brokerage & Info Day, Helsinki · 29 September 2026 — first pitch of the session: coordinators and pilot owners sought for the Digital Europe cyber call due in January 2027 and for Horizon Europe 2027. A pitch, not a project award.
• PKI Consortium PQC Conference, Amsterdam · 1–3 December 2026 — talk “PQC-ready is not PQC-verified” accepted by the programme committee; forthcoming. Conference page ↗
• IETF 126 side meeting · 20 July 2026 — contribution to “Remote Attestation Challenges for AI Agents.” The discussion converged on a workload-neutral boundary: RATS appraises relevant properties of an untrusted endpoint; delegation and action records remain separate, composable layers. This is a side-meeting outcome, not IETF consensus or a working-group decision. Speaker notes and figure →

Memberships & community

• Apply AI Alliance (European Commission) — member by self-registration in the Stakeholders' Catalogue, which is not publicly viewable.
• ETSI Ambassador programme — Anton Sokolov takes part in a personal, unpaid capacity (selected September 2026). Ambassadors do not speak for ETSI; this is not ETSI membership, a role of Tyche Institute, or an endorsement.
• PKI Consortium — member.
• Decentralized Identity Foundation (DIF) — Contributor.
• Agentic AI Foundation (AAIF), Linux Foundation — Associate Member with effect from 1 September 2026; one agreement, executed 4 September 2026, covering both Linux Foundation and AAIF Associate Membership. No vote, no board seat, no committee seat. Membership note →
• Trust over IP (ToIP), Linux Foundation — Contributor.
• Coalition for Secure AI (CoSAI), Workstream 3 (Risk & Governance) — contributor.
• W3C Credentials Community Group — participant (Community Group, not W3C Membership).
• OpenID Foundation DCP Working Group — participant under the contribution agreements (not OIDF membership).
• RIPE Atlas — public software probe 1016856 hosted on Tyche infrastructure, with a privacy-minimised Network Evidence Observatory.

Full status, including pending applications, is on the Memberships page.

Open research

Working papers and datasets are deposited on Zenodo under CC BY, with permanent DOIs. Selected outputs on AI governance, eIDAS / EUDI, and attestation:

• A 250-jurisdiction dataset of public AI-governance observability — 10.5281/zenodo.20999217.
• Sole Control for Machines — the EU remote-QSCD-management census and the reference implementation of the accountable agent authorisation model, data and code, v1.1 — 10.5281/zenodo.23321086 (first-wave snapshot 10.5281/zenodo.20990693).
• Have Your Say — stakeholder responses on EU digital-identity, trust-service and AI files, 2026 snapshot; the corpus behind the Cloud and AI Development Act position — 10.5281/zenodo.21970118.
• MachineMandate — a real-phone evidence bundle: bounded SD-JWT VC delegation to an autonomous agent in an EUDI reference wallet, with attested four-gate enforcement — 10.5281/zenodo.21229258.
• EATF-MultiRoot v0.2.3 — a reproducible verifier and 727-package corpus for AI-agent evidence, under review at SoftwareX (second revision submitted 5 October 2026) — 10.5281/zenodo.21874026.
• Project Veraison challenge-nonce freshness study — end-to-end appraisal of application-layer action evidence through TPM reference schemes — arXiv:2608.03534.
• EU Trusted List authorization half-lives — public data and reproducibility bundle supporting a time-to-event study of mutable trust registries — 10.5281/zenodo.21886352.

The full working-paper list, with DOIs, is on the Research page.

Public software

Source under github.com/tyche-institute:

• machine-mandate — an action-sealed authority gate for autonomous AI agents in the EU Digital Identity Wallet.
• aep-sandbox — an attackable red-team sandbox for Action Evidence Packages against a mandate-enforcing verifier.
• eatf-verifier — reference implementation and specification for cryptographically verifiable AI-agent self-attestation, with verify-aep-action, a drop-in CI check.
• time-travel-evidence-lab — an EU-wide census showing 43% of 4,815 trust-service entries drift from valid to revoked over time, across all 26 member-state lists.
• eatf-multiroot-softwarex — the reproducible SoftwareX artifact, manifest, verifier, and generated evaluation corpus for the v0.2.3 revision.
• aep-veraison-freshness — the reproduction package for the Project Veraison challenge-nonce freshness study.

Live tools & observatories

• AI-Governance Observability Index — ai-observability.eatf.eu.
• Agent attestation demonstrator — an agent acting under an executable mandate, with enforcement gates — agent.eatf.eu.

More on the Observatories page.

Peer review & service

Journal peer reviewer (2026): Journal of the Association for Information Science and Technology (JASIST), Sage Open (three reports), JeDEM — eJournal of eDemocracy and Open Government, and the Journal of Standardisation. Programme-committee member, 1st International Workshop on Explainable AI for Enterprise Architecture and Decision-Making (XAI-EADM) at CBI & EDOC 2026.

Unlike the rest of this page, these two have no public anchor: journal review assignments are confidential, and the workshop does not publish its programme committee. They are stated on our own record and cannot be verified from outside.