Skip to content
Tyche Institute

Field reports · 2026

Expeditions

A few times a year the institute's research leaves the desk and goes where the people are who will argue with it: a summer school, a standards workshop, a hackathon table, a national registry operator's office. An expedition bundles those stops into one journey, with study days in quiet places between them and a visit to a real laboratory or observatory wherever the road passes one. The point is not travel. It is that a verifier demonstrated to strangers gets broken by a stranger, and a question put to a panel gets an answer the literature does not contain.

Each report below is the public version of a longer internal account. People in private conversations are named by role, not by name; presenters and panellists at public events are named where they were public. Nothing here is a claim of endorsement by the institutions visited. Maps are drawn in an equidistant conic projection; straight lines join consecutive stations and are not the paths travelled.

Expedition 1 · 6–20 July 2026

Summer Bend: Brussels, Epirus, Lesvos, Vienna

15 days · 4 countries · 1 verifier broken by a participant · 1 Internet-Draft and 1 media type filed

Route map of the Summer Bend expedition: Tallinn, Brussels, Thessaloniki, Ioannina, Mytilene, Athens, Vienna and back to Tallinn, drawn in black on white
Route of the Summer Bend, from Tallinn and back. Coastlines: Natural Earth.
No. Dates Station What
2 7–8 July Brussels, Belgium Final conference of the ENFIELD network of excellence on trustworthy AI, Academy Palace
3 9 July Thessaloniki, Greece Arrival in the evening; by car west into Epirus
4 9–12 July Ioannina and the Pindus, Greece Lake Pamvotis, the citadel of Its Kale, the Vikos gorge; a research conversation; study days
5 12–17 July Mytilene, Lesvos, Greece ENFIELD summer school AICYBERSEC 2026 at the University of the Aegean (15–17 July): live verifier session on the last day
6 17–18 July Athens, Greece Transit
7 18–20 July Vienna, Austria IETF 126: hackathon table (18–19 July), side meeting on agent attestation (20 July)

Brussels. The ENFIELD network of excellence closed its Horizon Europe run on trustworthy AI with a final conference at the Academy Palace on Rue Ducale. The institute's question there was the one that runs through the whole year: what should an AI agent have to prove before it is allowed to act in someone's name? The answer we argued for is a mandate that is bounded and revocable, carried on the rails of the EU Digital Identity Wallet, with every action leaving signed evidence a stranger can verify without asking us.

Epirus. From Thessaloniki by car west into the Pindus, to Ioannina on Lake Pamvotis, and the next morning into the Vikos gorge, said to be the world's deepest for its width. Off every programme; the study days that an expedition needs.

Lesvos. The ENFIELD summer school AICYBERSEC 2026 in Mytilene, hosted by the University of the Aegean, with Anatolia across the strait. On the last day we opened the live verifier on a laptop and asked 5 researchers to break it rather than admire it. One of them took the public artefacts home, ran the replay, matched the manifest, and handed back the defect we had missed: two different attestation quotes travelling under the same filename. That is the outcome an expedition is for.

Vienna. IETF 126, the institute's first IETF meeting: 2 days at a hackathon table by the Stadtpark, where the evidence format was cross-verified against an open-source COSE provenance stack from Telefónica, with byte-identical canonicalisation and a negative vector failing closed. The same week the institute filed its first individual Internet-Draft, composing an Action Evidence Package with remote attestation under RATS, and the media type that IANA registered in August. On 20 July a side meeting on remote attestation for AI agents; the speaker notes are public.

What came of it. The replay defect from Lesvos was fixed and the verifier re-released. A freshness gap we had reported in an open-source attestation verifier got part of its fix in review, written by someone we had never met. The Internet-Draft reached its 6th revision by the end of August and became the companion of an IEEE Internet Computing article.

Expedition 2 · 9 September – 4 October 2026

North Bend: Reykjavík to the Haute-Provence Observatory

26 days · 7 countries · 11 stations · 6 talks · 1 pitch · 1 radio interview · 1 demonstration table

Route map of the North Bend expedition in two parts from Tallinn: Reykjavík, Enschede, Rome, Portovenere, Pavia, Porto Ronco, then Helsinki, Sophia Antipolis, Marseille and the Haute-Provence Observatory, drawn in black on white
Route of the North Bend, from Tallinn and back, twice. Solid line: part 1, 9–24 September. Dashed line: part 2, 25 September – 4 October. Coastlines: Natural Earth.
No. Dates Station What
1 9–14 September Reykjavík, Iceland 5 research conversations on trusted lists, population registers, web archives, research ethics and connectivity (10–11 September)
2 14–15 September Enschede, Netherlands XAI-EADM workshop at CBI & EDOC 2026: paper talk, 15 September
3 15–18 September Rome, Italy 3 workshop talks at ESORICS 2026 (TCI, STM, RAISE), 17–18 September
4 18–21 September Tuscany and Liguria, Italy Study and preparation in Portovenere
5 21 September Pavia, Italy First in-person meeting with the SAFE AI co-authors, University of Pavia
6 21–24 September Porto Ronco, Lago Maggiore, Switzerland Quiet study and preparation
7 25–28 September Tallinn, Estonia BSides Tallinn main-stage talk (25 September); Atlas radio interview, FORRT lightning talk, TalTech doctoral kick-off (28 September)
8 29 September Helsinki, Finland Connect4Cyber brokerage day: pitch and partner talks
9 30 September – 1 October Sophia Antipolis, France ETSI/CEN workshop on the EU Digital Identity Wallet, with a demonstration table in the breaks
10–11 1–4 October Marseille and the Haute-Provence Observatory, France 3 science visits: a round table on life elsewhere, a science night, the observatory's open day

The expedition had two parts and 5 results that matter most.

Reykjavík: what survives of a signed public record. 5 conversations: the operator of the national trusted list, a specialist at the population register, the chair of the research-ethics committee, the head of digital projects at the national library, which runs the web archive, and the chief executive of the submarine-cable operator. The trusted-list operator confirmed that the version number rises by 1 per release, which validates a denominator in our trusted-list history work, and that signer names in archived versions are public data. The lesson is about custody: a signed record is only as durable as the institution that keeps copies of it.

Enschede: a formal neighbour we did not know. At the XAI-EADM workshop at CBI & EDOC 2026 we presented carrier-bound evidence profiles for X-Road. The person who started the EDOC conference series pointed us to delegation chains in the enterprise language of the Open Distributed Processing reference model; within days he had extended his own verifier for multi-parent joins, and all 98 verdicts of our 12 MandateGraph test cases agreed with it. Our related work gained 2 neighbours it had missed.

Rome: when 2 verifiers disagree, the specification is the third judge. 3 talks at 3 ESORICS 2026 workshops at Sapienza University: brokered trust against single-root failure (TCI), the SD-JWT differential-fuzzing paper now published in LNCS (STM), and cross-verifier disagreement in AI-agent evidence over X-Road (RAISE). A conversation after the last talk put our IETF draft on the desk of 2 researchers at a Dutch applied-research organisation; nothing is agreed, and a second, independent implementation is what we hope for.

Pavia and the lakes. First in-person meeting with the co-authors of our work on safe-AI metrics at the University of Pavia, then 3 quiet days on Lago Maggiore. Study days are where talks get rebuilt and the next month's manuscripts begun.

Tallinn, Helsinki. The main stage of BSides Tallinn on 25 September, the largest room we have spoken in: recording. On 28 September the retraction research reached 2 outside audiences: an interview for Atlas, the science programme of Estonian public radio (broadcast 4 October, with an article on ERR News), and a lightning talk at FORRT's online conference on AI in metascience (recording). On 29 September, at the Connect4Cyber brokerage day in Helsinki, the institute pitched first in the session: coordinators and pilot owners wanted for the cyber calls of 2027.

Sophia Antipolis: the certification asymmetry, said aloud. At the ETSI/CEN workshop on the EU Digital Identity Wallet we ran a demonstration table in the breaks, with the mandate-and-evidence demonstration on a laptop. In the session on relying-party registration we asked the panel whether anyone checks that a relying party's software rejects a tampered credential. The answer: eIDAS requires certification of qualified providers, not of relying parties, and no one is responsible for the relying-party side. The same gap came up 4 more times in 2 days, from a legal expert's slide, from the speaker of ENISA, the EU Agency for Cybersecurity, and from the Commission's own conformance framework for the wallet, which tests wallet interfaces only. Field evidence for a manuscript in preparation; the limit stays that our census measured libraries, not deployed relying parties.

Provence: real science beside the road. A round table in Marseille on the possibility of life elsewhere, with the Jupiter Icy Moons Explorer and the exoplanet count on the slides; a science night where a researcher advised us to learn French, the fourth language recommended on this trip; and the open day of the Haute-Provence Observatory, in the dome of the 1.93 m telescope with which the first exoplanet around a Sun-like star was found in 1995. At the aerosol network's stand we asked what happens to field data when the calibration after deployment differs from the one before. They recalibrate. That is a measurement discipline with a chain of custody, and it is our subject in another language.

Epilogue · 8–11 October 2026

Brussels and Paris: a committee, a pendulum, a service of measurements

4 days · 1 working day · 1 committee seat taken up · 2 trains · 4 science visits

Route map of the epilogue: Tallinn, Helsinki, Brussels, Paris, Brussels, Tallinn, drawn in black on white
Route of the epilogue, Thursday evening to Sunday night. Coastlines: Natural Earth.
No. Dates Station What
2 8 October Helsinki, Finland Transit on the evening flight to Brussels
3 9 October Brussels, Belgium First Management Committee meeting of COST Action CA25178; evening train to Paris
4 10–11 October Paris, France Fête de la Science, last weekend: Jardin des Plantes, the Panthéon, Musée Curie, ESPCI, Institut d'Astrophysique de Paris
3, 1 11 October Brussels, then Tallinn Train to Brussels, evening flight to Tallinn

Friday, Brussels. The first Management Committee meeting of COST Action CA25178 on agentic physical AI, at the COST Association, 09:30 to 17:00. Officers elected, 5 working groups set up, and the institute's theme, evidence and conformance for the actions of embodied agents, placed as a cross-cutting task with 2 concrete pieces of work taken on in the room. The news item has the detail. Then the evening train to Paris.

Saturday, Paris. The last weekend of the Fête de la Science, spent in the Latin Quarter. The Jardin des Plantes first, with the plaque on the house where Becquerel found radioactivity in 1896. Then the Panthéon, where Foucault's pendulum was photographed twice, 43 minutes apart: at the latitude of Paris the plane of swing should have turned about 8 degrees between the 2 pictures. At the Musée Curie, a panel on the laboratory's service des mesures: for decades it measured samples of radium for industry and medicine and issued a numbered certificate for each one, thousands in all, much of the work done by women. A trusted laboratory, a numbered certificate, a signature that industry and medicine accepted: the institute's subject, a century early. In the same room a particle-physics laboratory from Caen assembled a cloud chamber with the public, from Wilson's question about how clouds form to a neutrino telescope on the floor of the Mediterranean.

At the open day of ESPCI Paris, a public lecture by a physicist of the Institut Langevin on walking droplets: a drop that bounces on a vibrating bath and is pushed along by its own waves, 2 drops that talk to each other through the surface, a time mirror on water. We asked, in English, whether the drops reach de Broglie's pilot wave and non-locality. The honest answer was that non-locality is what the droplets cannot do. The lecture produced a notebook, a small simulation and a brainstorm at home; whether it produces more is a later decision.

Sunday, Paris. The Institut d'Astrophysique de Paris: the Euclid stand, where we asked about weak lensing and the dark-matter map, and a public talk by the institute's director on hunting neutrinos to read the violent universe, including the radio-antenna array of the GRAND collaboration and a timeline of how geopolitics shaped where the experiment could be built. A 5-minute conversation afterwards about data provenance and integrity in a large distributed instrument ended with a request for a one-page note, which went out the next day. Then the train to Brussels and the evening flight to Tallinn.

Three laboratories on the weekend's plan were not visited; the plan was a menu, not a timetable. Nothing in this epilogue was recorded for publication, and no slide of another presenter is reproduced.

Expedition 3 · 30 October – 17 November 2026

West Bend: San Francisco Bay to Death Valley

19 days · 2 states and Tallinn · 13 stations · 2 sessions convened · 1 demonstration table · 1 hackathon · 1 archive

Route map of the West Bend expedition in California and Nevada: San Francisco, Berkeley, Sunnyvale, Mountain View, Stanford and SLAC around the Bay, then Big Sur, Sequoia National Park, Death Valley, Las Vegas and Pismo Beach, with Tallinn off the sheet, drawn in black on white
Route of the West Bend in California and Nevada. Inset, solid line: the programme days around San Francisco Bay. Dashed line: the road south, 6–12 November. Small circles are nights and sights between the stations. The flights are not drawn; station 13, in Tallinn, lies off the sheet, and the disc at the upper right stands for it. Coastline, state borders and lakes: Natural Earth.
No. Dates Station What
1 30 October – 2 November, 13–15 November San Francisco, California Arrival on 30 October; the base for both weeks by the Bay
2 2 November Berkeley, California A conversation at the Center for Human-Compatible AI, University of California, Berkeley
3 2 November Sunnyvale, California A call with the maintainers of OSPREY at Berkeley Lab; the OpenID Foundation workshop
4 3–6 November Mountain View, California Internet Identity Workshop #43 at the Computer History Museum: 2 sessions convened and demonstration table 7; Agentic Internet Workshop #3
5 4–5 and 12–13 November Stanford, California Hoover Institution Library & Archives; the Arrow Lecture; the showcase of the McCoy Family Center for Ethics in Society
6 7 November Big Sur, California Highway 1 and the elephant-seal rookery at Piedras Blancas: the road south begins
7 8 November Sequoia National Park, California Grant Grove, the Giant Forest, Moro Rock
8 9–10 November Death Valley, California The Mesquite Flat dunes, a night under the stars, dawn at Zabriskie Point
9 10 November Las Vegas, Nevada Unplanned: 2 hours that became 5
10 11–12 November Pismo Beach, California The Pacific again; a desk day
11 13 November SLAC National Accelerator Laboratory, Menlo Park, California A public tour of the accelerator laboratory
12 14–15 November San Francisco, California IETF 127 Hackathon: a sealed test pack and a blind round; version 01 of an Internet-Draft finished at the table
13 17 November Tallinn, Estonia The first day of a course on European Union project proposals at the Information System Authority (RIA)

The third bend of the year, and the first outside Europe. Work on Estonia and the European Union went, on purpose, to a country that has no X-Road and no European Digital Identity Wallet: the contrast is the material.

Berkeley and Sunnyvale, 2 November. A conversation at the Center for Human-Compatible AI was followed that afternoon by a page of ours, “Who acted. Who could check. By what rule.” It traces the fork between European and American conformity to 1866, when 2 bodies were founded to stop boilers exploding: in Mannheim a society of boiler owners for inspection, which grew into Germany's technical inspection associations (TÜV), and in Hartford an insurer that sold cover only together with its own inspection. Europe went on to name the body that checks before the market; America, who pays after the failure. In Sunnyvale, at the OpenID Foundation's workshop, a question from the floor became a proposal within 2 hours: a “receipt” obligation type for AuthZEN, the Foundation's authorisation interface, so that an agent can be asked to prove afterwards that it did exactly what was permitted. It is proposal #795 in the working group's repository, open for discussion and not adopted.

OSPREY. The same day, a call with the Accelerator Physics group of the Advanced Light Source at Lawrence Berkeley National Laboratory (Berkeley Lab), which maintains OSPREY, an open-source agentic interface for accelerator control. That evening we sent 1 page and 12 scenario files run against its public code. The next day the maintainer filed 2 findings, on what the agent record keeps and when a write asks for approval, as public issues #2336 and #2337, with credit. The software and any fixes are its maintainers' work. It is not an agreement and not a partnership.

Mountain View: a mandate, not an identity. At Internet Identity Workshop #43, an unconference at the Computer History Museum, we convened 2 sessions. The first, “Who Authorized This Agent Action?”, fell on 3 November, Election Day in the United States. One slide listed what European law already asks for when software acts, from the AI Act and the Product Liability Directive to the harmonised standards under way and eIDAS 2, and asked where the United States stands. Neither that room nor the next named a federal counterpart; both pointed to the states. That is a scoping observation from 2 rooms, not a survey of American law. The second session came from the first day, when nobody in the room had heard of Estonia's plan of June to give AI agents identity codes, withdrawn by September: “ID codes for AI agents? Estonia said no. So what should an agent carry?” The session's posted notes give its result as “verifiable mandate rather than verifiable identity”. At table 7 of the Speed Demo Hour the questions led to a design decision taken the next morning: the verifier stands outside the agent. The table also won a prize, a LEGO Technic Bush Plane of 333 pieces. On 5 November we co-led a thread at the Interop Day of Agentic Internet Workshop #3, where revocation was, once more, the part still unsolved.

Stanford: a paper ancestor, and a limit. On 5 November, a first working day in an archive, in the reading room of the Hoover Institution Library & Archives: Soviet security files on Estonians in the Herbert Romerstein collection; letters that Toomas Hendrik Ilves wrote in 1984–85 on the letterhead of Radio Free Europe, in the Wayne Holder papers; and, on microfilm, read on a second visit on 13 November, the circulars in which the Russian secret police abroad told its offices which codes were compromised and withdrawn from use, 1906–1914. They are a paper ancestor of the revocation list. That evening Betsey Stevenson of the University of Michigan gave the public Arrow Lecture, “What Do We Owe Each Other in the Age of AI?”. On 12 November, at the showcase of the McCoy Family Center for Ethics in Society, we asked whether AI can be given limits the way colour television sets were, after some of them were found in 1967 to emit X-rays. Congress did not ban television. Under the Radiation Control for Health and Safety Act of 1968 it set a limit, 0.5 milliroentgens per hour at 5 centimetres from the set, tested with the controls at their worst. The conclusion is ours: a limit needs a quantity, a unit, an instrument and a test condition, and for most harms of AI nobody has the instrument yet.

The road south, 6–12 November. A week of our own, with the desk along. It began with a guest lesson by video, “Numbers wear costumes”, for a high-school statistics class in the United States. Then Highway 1 to Big Sur and the elephant seals of Piedras Blancas; the sequoias of Kings Canyon and Sequoia National Parks, where the General Sherman Tree is by volume the largest tree on Earth; the dunes of Death Valley. Death Valley is an International Dark Sky Park, and on a night without a moon a planetarium app named the Andromeda galaxy overhead; 5 nights later the San Francisco Symphony played Augusta Holmès's “Andromède”. Las Vegas was on no plan: an institute named after the goddess of fortune gave the city 2 hours, which became 5, and placed no bet. Then the Pacific again at Pismo Beach. 3 manuscripts left for journals from desks along the road.

SLAC: an access rule read as a conformity regime. On 13 November a public tour of SLAC National Accelerator Laboratory, joined at the door after both tours of the day had sold out: a bus along the 3.2 km Klystron Gallery and into an experimental hall of the Linac Coherent Light Source, which the laboratory describes as the world's first hard X-ray free-electron laser. The rule for the Department of Energy's user facilities has the shape of a conformity scheme: open to any interested user regardless of nationality or affiliation, resources allocated by merit review of the proposal, no fee for work whose results are published in the open literature, and the full cost for proprietary work. At the end of the tour we spoke about the provenance of scientific data and promised a page; it went out the next morning. A public tour is not an invitation, and nothing beyond it is arranged.

San Francisco: strangers can recompute it. At the IETF 127 Hackathon, on 14 and 15 November, we brought 2 projects. For the Action Evidence Package we published a test pack of 115 files, sealed with a hash and timestamps from 2 timestamping authorities before the hackathon opened: the text of the format, 22 test vectors and the rules of a blind round, in which participants get the text and the vectors and never our code. The round is open until 22 November, and what we expected of each vector is published by 29 November; until then we report no result. A verifier in Go written by Abdel Fane of OpenA2A led, on our side, to 2 public fixes and to release 0.7.4 of our verifier on 15 November. Version 01 of our Internet-Draft “Deterministic Encodings for COSE” was finished at the table, with 41 test vectors in place of 33. A checker for our draft, written from the text with an AI assistant, and reviewed and run by Laurent Lemonnier, passed 41 of 41 test vectors. On 17 November the draft was presented at the COSE working group session, by Steven Mih in the room and remotely from Tallinn; the discussion continues on the mailing lists.

The strongest objection of the trip, as we state it: an agent is just software, and between cooperating organisations isolated logs and internal controls are enough. Our answer is that the objection holds for cooperation and fails for dispute. When the parties do not trust each other, a log that its owner controls can show one history to one reader and another to the next, and preventing that takes a transparency log with witnesses.

Tallinn, 17 November. The flight landed after midnight, and the last day was a classroom: the first day of a course on preparing European Union project proposals at the Information System Authority (RIA). In California many halls carry the names of those who paid for them; in Europe research is called for from above, topic by topic, and answered by consortia formed across borders. A small institute has to learn both grammars.

What came of it. A proposal at the OpenID Foundation; 2 findings in OSPREY's public tracker; release 0.7.4 of the verifier; version 01 of the deterministic-encodings draft, presented at the IETF; a blind round that reports after 29 November; and 2 historical cases for the conformity argument, the steam boiler and the television set. Around the programme the desk kept moving: on 4 November our article on AI export controls was published in Frontiers in Political Science.

Private conversations are reported by their conclusions, without attribution. The workshop sessions are reported from their posted notes, in which speakers from the floor are not named.

How to read these reports

Every stop above has a longer internal account with sources, timings and the names of the people we spoke to. The public version keeps what a reader can check: programmes, recordings, published papers and registries, each linked from the track record and the news pages. Where a conversation was private, the report keeps the conclusion and drops the attribution.