Decisions

Governance Predicted Fraud Better Than Character

Founder-controlled boards were 88 percent likelier to face fraud. Governance beat founder traits as a predictor, and hot-market first rounds carry the marker.

DecisionsBy Servola Tech Desk2026-07-316 min read

AI-assisted, edited by humans. | Editor-in-Chief: Leon Soliman | Editorial standards

Governance Predicted Fraud Better Than Character

Key takeaways

Startups with founder-controlled boards were 88 percent more likely to be involved in fraud than those with investor-controlled or shared-control boards, and governance variables predicted fraud better than any founder characteristic; the strongest diligence signal on a venture-backed supplier is a board composition you can read in ten minutes, not a reference call about the founder.
Initial rounds raised in hot market conditions were associated with higher later fraud, which places the marker directly on the cohort being funded through the current artificial intelligence cycle; if you are evaluating a supplier that raised its first money in the last two years, treat the funding conditions as part of the risk picture.
Detected fraud reached 1.85 percent of venture-backed firms that raised at least 10 million dollars, rising from 0.11 percent in 2003 to 0.67 percent in 2021, and researchers found fraudulent founders went on to raise again unharmed; where the penalty channel is weak, a detection rate is a floor rather than an estimate.
Deception escalates with the gap between promised and actual performance through three stages, from lying about traction, to manufacturing evidence for the lie, to fabricating capability outright; the lever that stops the climb is resetting the expectation early, so treat a missed number as a moment to renegotiate rather than to work harder in silence.

Two research teams, two continents, one finding

Tim Weiss at Imperial College London and Nevena Radoynovska at Emlyon Business School in Lyon spent their study reading court records: the files of Silicon Valley ventures and the founders prosecuted for fraud between 2000 and 2023. Their paper, Criminal Deception in Silicon Valley, was published online in Organization Science on 29 June 2026.

A second team reached the same neighbourhood from the opposite direction. Alexander Dyck, Freda Fang, Camille Hebert and Ting Xu, working at the Rotman School at the University of Toronto, assembled what their National Bureau of Economic Research working paper describes as the first dataset of venture fraud cases, covering 654 United States venture-backed startups. It was issued as working paper 34868 in February 2026, drawing on regulatory and prosecutorial actions, shareholder suits and reported cases.

One study is qualitative and reads how deception was actually carried out. The other is quantitative and asks what predicts it. They were built independently, with different methods, and they converge on a conclusion that is not where most people look.

The ladder a founder climbs

Weiss and Radoynovska describe deception as a ladder with three rungs, not a single act. The first is surface facading: telling investors the company is doing better than it is. This is the common one, and it lives in the early pitch, where a founder is selling a vision and the distance between the story and the ledger is treated as ordinary optimism.

The second rung is reinforced facading, where the founder manufactures evidence to support the original claim. The third is deep facading, where the technology itself is fabricated and a parallel version of the company is constructed and maintained for outsiders to inspect.

The mechanism matters more than the taxonomy. The researchers tie which rung a founder reaches to the size of the gap between the performance investors expect and the performance the company is delivering. Deception escalates because the gap widens, not because a person changes. That reframes the whole problem: the input that drives the climb is the expectation that was set, and expectations are something a board, an investor and a customer all help to set.

The number that actually predicts it

The Toronto team's central result is that governance characteristics, rather than founder traits, are the strongest predictors of fraud. The single sharpest variable is who controls the board. Startups with founder-controlled boards showed an 88 percent higher likelihood of fraud than those with investor-controlled or shared-control boards. That figure has been circulating in shorthand as twice as likely, which overstates it; the published number is 88 percent, and the difference is worth keeping straight if you intend to use it as a threshold.

Three further conditions travel with elevated fraud: founder-friendly contract terms, complex capitalisation tables, and initial rounds raised in hot market conditions. None of these is a statement about anybody's honesty. They are structural features visible from outside the company, which is precisely what makes them useful.

On prevalence, detected fraud reached 1.85 percent of venture-backed firms founded since 2000 that raised at least 10 million dollars, and the ongoing rate climbed from 0.11 percent in 2003 to 0.67 percent in 2021. Among newly public companies, venture-backed firms were 54 percent more likely to face fraud litigation than comparable firms that had not taken venture money.

Detected is doing a lot of work in that sentence

Both teams land on the same uncomfortable observation about consequences. The Toronto research found little evidence that past fraud allegations stopped founders from raising again, and describes fraudulent entrepreneurs continuing to found new venture-backed startups unharmed, which it reads as weak market discipline. The authors put the conclusion plainly: the modern venture model itself may be creating conditions that make fraud easier to commit, harder to detect, and less likely to be punished.

Follow that through and the headline prevalence number changes character. A rate of 1.85 percent counts cases that surfaced through prosecutors, regulators, shareholder litigation or the press. If the channels that surface and punish fraud are themselves weak, the measured rate is a floor rather than an estimate, and the true figure sits somewhere above it by an unknown margin.

The hot-market finding is the part that is live right now rather than historical. Companies whose first round was raised into an overheated market with weak oversight were more likely to end up in a fraud case later. The current artificial intelligence funding cycle is producing exactly that condition at scale, which means the marker is being written onto a cohort that is being funded this year and will be selling to European buyers for the rest of the decade.

What to do with this on Monday

First, read the board before you read the founder. When you are about to make a venture-backed company a material supplier, an integration partner or an acquisition target, establish who controls its board. Founder control is the 88 percent marker, and it is a fact rather than a judgement. In the United Kingdom, Companies House gives you the directors of a domestic supplier for nothing; for a United States vendor you will have to ask for board composition in the diligence pack, and a refusal is itself information.

Second, insist on evidence you choose rather than evidence you are handed. The ladder tells you that manufactured proof is the second rung, so a curated demonstration and a reference customer selected for you sit exactly where a company under pressure would place them. Ask to speak to a customer you picked from a list, ask for the system to be exercised on your data, and treat a polished artefact as the weakest form of evidence rather than the strongest.

Third, apply the finding to your own company if it raised into this market. The escalation is driven by the gap between the promise and the position, so the protective action is procedural: bring a missed number to the board in the quarter it is missed, with a revised plan attached. The failure mode the research describes does not begin with dishonesty. It begins with a founder deciding to close the gap quietly before anyone notices.

Frequently asked questions

Does this mean venture-backed companies are untrustworthy?

No, and the base rates say so. Detected fraud reached 1.85 percent of venture-backed firms that raised at least 10 million dollars, which means the overwhelming majority of the population is not involved in anything of the kind. The useful claim is narrower and more actionable: within that population, structural features predict the risk better than impressions of the people, so diligence effort is better spent on governance than on character assessment.

Why would a founder-controlled board matter so much?

The research does not present it as a moral fact about founders. A board that the founder controls is a board that cannot compel disclosure, replace management, or force a reset when performance diverges from the plan. Since the studies tie escalating deception to a widening gap between expectation and reality, the governance structure that determines whether that gap gets confronted early is exactly the variable you would expect to matter.

We are a customer, not an investor. Is this relevant to us?

It is arguably more relevant. An investor can diversify across a portfolio and absorb one failure. A customer who has built a process on a single supplier cannot, and the cases in these datasets include fabricated technical capability, which is the failure mode that reaches a buyer directly. Board composition, contract terms and the market conditions of the first round are all available to you before you sign.

The instinct in diligence is to study the person, because fraud feels like a question of character. Two independent teams looking at the same two decades found that the structure around the person predicted it better. Board control is a ten-minute check, and on this evidence it outperforms every hour you spend forming an impression of a founder across a table.

Venture capitalCorporate governanceFraudDue diligenceBoard compositionOrganization Science

Nvidia and Samsung Pooled Their Materials Data

CuspAI confirmed a 450 million dollar Series B and launched the AI Materials Foundry with more than 45 founding partners. The consortium, not the round, is what changes your sourcing position.

5 min read

Capital

Germany's Record Seed Bets Robots Need Data

Munich's Microagi raised 55 million dollars, Germany's largest-ever seed round, for a platform that trains factory robots on a plant's own data. Why investors decided the robot bottleneck is data, not hardware, and what that means for European owners.

4 min read

Law & Regulation

The EA Buyout Now Hinges on Brussels

The largest take-private in history has cleared EA shareholders and US antitrust. The last real gate is the EU - and not classic merger review, but the Foreign Subsidies Regulation, facing its first blockbuster test.

3 min read

← All articles