When machines decide, who keeps it honest?

In short: “Prevention costs a couple of billion, and you cannot sell that with a value proposition.” Benjamin van Rooij’s keynote argued that organisational disasters are structural rather than the work of one bad apple, and three panels carried that into cyber resilience, disinformation and the boardroom. Simone Honig of NautaDutilh hosted and opened with five insights from criminal investigations; Casper Roerade set the geopolitical scene. The fuller account is below.

The 10th Week of Integrity reached its seminar on a Thursday afternoon at NautaDutilh in Amsterdam, and the programme did something slightly unusual for an event about ethics and technology. It refused to be comforting. The keynote told the room that most of what they do for a living treats the symptom rather than the disease. The panels that followed, on cyber resilience, on public discourse, and on the boardroom, kept circling the same stubborn point: the tools are new, the temptations are old, and the responsibility has never moved off a human desk.

Before the keynote: two warm-ups that were not warm-ups

The afternoon opened wide, with a brisk tour of the world the compliance officer actually operates in. Casper Roerade of evofenedex set the economic weather, and it was not mild. China now accounts for something like a third of global manufacturing, nine in ten solar panels, most of the world’s batteries, electric vehicles and rare-earth refining, and yet the picture underneath is of structural stress rather than strength: consumer confidence negative since 2021, fixed investment falling, total non-financial debt around three times GDP, and local governments watching their land-sale income collapse. Export muscle, he suggested, is the flip side of domestic stagnation. Add a volatile Strait of Hormuz, oil stubbornly high on fear rather than fundamentals, and a US-China race for artificial intelligence in which the Americans chase the strongest frontier model and the Chinese chase return on investment, and you have the backdrop against which every “Thursday afternoon decision” now gets made.

That phrase came from Simone Honig of NautaDutilh, the seminar’s host, who followed with five insights from the far side of criminal investigations, and it was the sharpest ten minutes of scene-setting imaginable. How, the speaker asked, do you defend a decision years later, when a prosecutor reads your emails already knowing how the story ends? Watch out for hindsight bias. Mind your minutes, because board minutes are the documents most often used against companies, and investigators read them for the dissenting voice that got softened and the concern that quietly disappeared between drafts. Remember that knowledge creates responsibility under Dutch law, so “putting someone on CC is a small step, but in an investigation it becomes a very big fact.” Treat AI meeting transcripts with care, because “AI does not pick up sarcasm,” and a joke in a meeting can be frozen into evidence; whether you record and retain is a governance decision, not an IT setting. And accept that the first signal of trouble is almost always already inside the organisation. The problem is rarely that nobody knew. It is that the signal did not travel.

The keynote: how organisations go bad

Then Benjamin van Rooij of the University of Amsterdam stood up and, by his own cheerful admission, declined to sell anyone a quick fix. “I’m a bit of a weird guy,” he warned. “I’m not the sort of guy that comes in and says, do some nudging here and there and you’ll get this. That’s just not me.”

His argument, built with Nicholas Lord of the University of Manchester, starts from a distinction most compliance functions blur. There is compliance as liability management, the programmes and policies that keep the organisation defensible, and there is the genuine prevention of harm, which is a different and far more expensive thing. “A compliance programme costs a couple of million,” he said. “Prevention costs a couple of billion, and you cannot sell that with a value proposition.” The honesty was the point.

To work out what actually causes catastrophe, he had studied around twenty-eight of them, public and private, using only public information: Boeing, BP, Enron, Volkswagen, Wells Fargo, Purdue Pharma, Wirecard, the UK Post Office, USA Gymnastics, the Catholic Church, Meta, OpenAI and more. Across all of them he found not a rogue individual but a recurring organisational pattern, five mechanisms that turn ordinary institutions toxic. Goals set unrealistically high, so that strain and incentives push people past the line. Dissent that gets silenced, through groupthink, habituation, retaliation or the active organising of silence. Opportunity for harm that the organisation itself creates through scale, access and the impunity that comes with power. The slow normalisation of deviance, where bad behaviour meets no response and quietly becomes the way things are done. And finally the denial of responsibility, the neutralising, blame-shifting and attacking of accusers that arrives once the harm is exposed. Tone at the top and the wider economic and political context sit around all five.

He was careful about what this is and is not. “Please don’t see these as predictors,” he said. “This is not a scientific study where I can say something is causal.” They are patterns to recognise, prompts to ask “do I see something like this in my own organisation?”

The examples did the persuading. Tobacco companies, he noted, taught their own employees how to use industry-funded doubt science to discuss the cancer link with family and friends, so staff could “still feel good at work.” He reached back to BP’s “Beyond Petroleum” era, when the messaging at the top sounded, he said, “like Greenpeace, literally,” while people lower down knew there was no safety investment and accidents were frequent. He made it personal: his father, he told the room, was the second lawyer at Shell, and Shell people “howled with laughter” when Beyond Petroleum launched, “because we knew what their operations were like.” His lesson was not cynicism but modesty. Close the gap between the tone at the top and what he called the smell at the bottom. Be less ambitious and more realistic about ethics, so that people actually believe you. He pointed to the FAA signing off on Boeing while its agents’ annual reviews depended on whether Boeing hit production targets, and to whistleblowers inside the EPA sidelined after a phone call. The throughline: harm normalises, and organisations are extraordinary multipliers, for good and for bad.

Panel one: cyber, supply chain and the things that break

Moderated by NautaDutilh’s Danique Knibbeler, the first panel turned to operational integrity, and Ty Francis gave it its best images. We are, he said, “taking Formula One engines developed in the last six months and dropping them into old 1990s cars.” The ecosystem cannot hold the power. And yet, he added, “brakes are there to help the drivers go faster.” He walked through the new Dutch Cyber Security Act and its three clocks, twenty-four hours to alert the authorities, seventy-two for a full overview, around four weeks for root-cause analysis, “ticking whether the board knows or not.” His governance test for any AI system was a single question: “who can switch it off? If you can’t point to it now, that’s not governance, that’s just wallpaper.”

Genna Glaser of Braskem pushed back usefully on the seventy-two-hour framing. In petrochemicals, she said, real response is measured in minutes within the first hour, operators are drilled for life-critical scenarios, and there is always a human validation step, “never singularly AI, never a single individual.” Louis van den Bogaard of ASML located the biggest risk in the supply chain, where the weakest link sits, and made the case for the boring work: tabletop exercises that expose the gap you never planned for, such as discovering that the person who runs crisis communications is also on the crisis team and therefore “wasn’t there.” He also warned against drowning people in training until their “information budget” runs out, when “everything is important, but nothing is anymore.”

Susan du Becker of Microsoft was the bluntest. Audiences have become desensitised to breaches, she said, so what matters now is the reaction. She contrasted Johnson and Johnson in the Tylenol crisis, who pulled everything, apologised and “fell on their sword,” with companies that tried to hide and were destroyed. On a recent Dutch data breach she praised the sites that “went dark immediately” and were transparent. Her advice on artificial intelligence was quotable and correct: “start with governance, and please don’t give it to Frank in IT to run.” Build a centre of excellence, because “AI is a very heavy tool, you can do a lot of damage.”

The panel closed on the scenario everyone fears, a deepfaked chief financial officer demanding four million euros in thirty seconds. Du Becker’s answer was a laugh and a refusal: “Are you insane? We invented Teams calls.” Francis made it cultural, citing a Ferrari employee who foiled a CEO-impersonation scam by asking the caller what book he had last recommended. The real control, both agreed, is the psychological safety to say no, not another policy. And on scaling AI, Francis offered a warning for the policy writers: agents cannot read “use your own judgment.” He had run nineteen thousand reports through AI in seconds, only to find the tool keyed on the word “policy” and sailed past anything called a procedure, a manual or a code. “We are going to have to start writing our policies in a different way,” one version for humans and one for the machines.

Panel two: disinformation and the integrity of public discourse

Joris Luyendijk, who told the room he had handed in his press card to campaign full time for the defence of Europe, moderated a panel that ranged from newsrooms to geopolitics. Natali Helberger of the University of Amsterdam brought the research and the best rebuttal of the day. Citing a six-country study, she noted that people increasingly use generative AI for information yet distrust journalism that uses AI, and that trust in institutions falls as we lose the ability to tell human from machine. She refused the comfortable metaphor. “It’s not a black box. People are building this. If I don’t understand it, why do they unleash it on all of us?” The risk-only framing, she argued, is the easy half; the neglected half is using AI for human flourishing, such as the “modular journalism” that gets young people reading the news again. And with European universities “98% dependent on US technology,” her closing line doubled as a plea: “Europe, be a bit more self-confident, please.”

Luyendijk supplied the provocations, describing Europeans as “cappuccino-drinking, part-time-working, surrender monkeys” who are in fact “much tougher, we just don’t have any eloquence,” and insisting that the moral content of AI is ours: “people build it, people use it.” His line on the news business belongs on a wall: “information is not shampoo.” Vivianne Weehuizen of Fugro grounded the discussion in commercial reality, admitting the company had removed its diversity and inclusion statement from its website to win work in the Americas, even though the commitment “still runs through the veins of the company.” Her chief executive’s maxim, “norms are good, but trust is better,” framed the daily dilemmas, including a vessel currently working in Venezuela. Asked to name an integrity “North Star” company, she answered plainly, “none.” Otto van der Harst of the Stichting Reclame Code completed the panel. The audience questions were among the liveliest of the day, from a reminder that Huxley’s Brave New World was Nobel-nominated nine times, to a Chinese attendee offering the Dutch word hoogmoed and noting that in China AI is framed not as a threat but as a new industrial revolution, to a challenge to the tired slogan that “the US innovates, Europe regulates,” which drew Helberger’s firm reply that regulation has always enabled innovation.

Closing panel: integrity in the boardroom

Vera Cherepanova of Boards of the Future moderated the final session, and offered a technique worth stealing: when a board gets defensive about ethics, stop asking them to be ethical and ask instead “how can we make this ten per cent more ethical?” Geerte Hesen of Ferrovial reframed the central question from “can we do it” to “should we,” with a simple public test, “will we be comfortable explaining this?” She gave the example of resettling communities for road projects in Peru humanely even where the law grants the right of way. Fabia Tetteroo-Bueno, after a long career at Philips and now a non-executive director, offered three quick filters, is it illegal, does it hit reputation, does it hit quality, and a warning about artificial intelligence that landed cleanly: “AI doesn’t go to jail. Don’t outsource your brain.”

They borrowed Michele Wucker’s image of the grey rhino, the obvious danger that everyone sees charging and nobody stops. For Hesen the grey rhino is middle management, the layer where integrity messaging so often breaks down. For Tetteroo-Bueno it is the short-term miracle result: “three hundred million, let’s open the champagne. First let’s check how you got that.” She closed with a plea on diversity, now so contested that companies “even hide that they’re doing things on it,” reminding the room that one in five women is misdiagnosed because the research was done on men. Cherepanova reframed the whole debate as a search for “optimal diversity,” demographic and cognitive at once.

The thread that held

What connected the keynote to the panels was not a technology. It was a refusal to let anyone hide behind one. Van Rooij’s toxic organisations, Francis’s switch-off question, du Becker’s centre of excellence, Helberger’s insistence that someone built the black box, Tetteroo-Bueno’s reminder that the machine cannot be jailed, all of it pointed back to people and the systems they design around themselves. Leading with integrity in a digital age, the week’s theme, turned out to mean something quite old-fashioned. Decide as though you will have to explain it, keep a human in the loop, and make sure the signal travels.

The Week of Integrity closes with a lunchtime conversation on 30 October, “Is it even AI?”, and a spotlight on International Anti-Corruption Day on 9 December.