The accusation that cracked the silence

Scott Stevenson, co-founder and CEO of the legal AI startup Spellbook, posted on X last month calling out what he described as a “huge scam” running through the AI startup ecosystem. His accusation was direct: AI companies are manufacturing the appearance of record-breaking growth by inflating a metric called ARR — annual recurring revenue — and the biggest venture capital funds in the world are helping them do it.

“The reason many AI startups are crushing revenue records is because they are using a dishonest metric,” Stevenson wrote. “The biggest funds in the world are supporting this and misleading journalists for PR coverage.”

That last line is the one that matters most. Stevenson did not describe VCs as bystanders watching founders exaggerate their numbers. He described them as active participants — coordinating with startups to feed distorted figures to journalists and generate favorable press coverage. This is not a story about a few ambitious founders stretching the truth on a pitch deck. It is an accusation that institutional capital is running a coordinated narrative operation.

Stevenson acknowledged he was not the first person to raise this alarm. Multiple news reports and social media posts had already documented what insiders were calling ARR “shenanigans” before his post went public. That prior criticism had not produced meaningful accountability or changed behavior across the industry. The pattern continued, which is precisely why Stevenson’s willingness to go on record under his own name — as an active founder inside the ecosystem — broke through in a way earlier warnings had not.

Founders do not typically accuse their investors and peers of running a scam in public. The social and financial costs are real. That Stevenson did it anyway signals that the gap between the numbers being reported and the underlying business reality had grown wide enough that silence felt like complicity.

What ‘ARR’ actually means — and what AI startups are making it mean

Annual Recurring Revenue was built for one specific business model: subscription SaaS. A customer signs a contract, pays a predictable amount every year, and renews. You take that contracted value, sum it across active customers, and you have ARR. The metric works because the underlying revenue is locked in, predictable, and genuinely recurring. That precision is the entire point.

AI startups are dismantling that precision systematically. The most common manipulation is annualization: a startup closes a $50,000 pilot in October, multiplies by 12, and announces $600,000 in ARR. The pilot has no renewal clause. It may never be paid in full. It doesn’t matter — the number is already in the press release. Scott Stevenson, co-founder and CEO of legal AI startup Spellbook, called this out directly on X, describing it as a “huge scam” backed by some of the biggest funds in venture capital. His accusation wasn’t aimed at fringe operators. He pointed at the institutional layer enabling and amplifying these figures.

One-time enterprise deals get the same treatment. API usage fees — which fluctuate month to month based on consumption with no contractual floor — get annualized as if they represent durable commitments. Consumption-based pricing is the dominant model across AI infrastructure and many AI application businesses, and it is structurally incompatible with ARR as a metric. When a customer’s usage spikes during a product evaluation and then drops, annualizing the peak produces a number that describes a business that does not exist.

This is where most coverage stops short. The ARR inflation story gets framed as a ethics problem — a few dishonest founders gaming a metric. The structural issue runs deeper. ARR was never designed for usage-based or consumption-based revenue models. Applying it to AI businesses that price on API calls, tokens processed, or seats that churn freely isn’t just aggressive accounting. It’s using the wrong instrument entirely, then reporting the reading as fact.

How VCs became co-architects of the inflation machine

Venture capitalists don’t just tolerate inflated ARR — they manufacture it. When a portfolio company announces a headline revenue number, that figure flows directly into the VC’s own valuation models. A startup claiming $50 million ARR commands a higher paper valuation than one reporting $20 million, which means the fund’s LP statements look stronger, the next fundraise gets easier, and the partners’ carry inches upward. The financial incentive to let aggressive accounting slide is not incidental. It is structural.

The PR machinery makes the problem worse. VCs routinely brief journalists on portfolio company milestones, positioning these figures as independent data points rather than what they actually are: investor-sponsored narratives. When TechCrunch or Bloomberg reports that a startup has “surpassed $X million in ARR,” readers assume some form of verification occurred. In most cases, the number came from the founder’s mouth, passed through the VC’s press relations contact, and landed in the article without audit. The VC has effectively laundered promotional copy into editorial coverage.

Scott Stevenson, co-founder and CEO of the legal AI startup Spellbook, called this out directly, writing on X that “the biggest funds in the world are supporting this and misleading journalists for PR coverage.” His framing matters: he did not accuse a rogue startup of bending the rules. He accused the largest, most established players in venture capital of actively enabling the game.

The feedback loop that follows is self-reinforcing and fast. Inflated ARR drives a higher valuation in the next funding round. The larger round funds a bigger sales and marketing operation. That operation generates the next wave of usage activity — however loosely defined — which produces a new, even larger ARR figure to announce. Each cycle pulls in fresh capital from investors who see only the headline number and benchmark it against the previous one. By the time the underlying economics surface, the early investors have often already marked up their positions and moved on. The ones left holding the exposure are the later-stage funds, the growth equity investors, and eventually the public markets or acquirers who discover the gap between reported ARR and actual, durable revenue.

Why journalists and the media ecosystem keep getting played

Tech journalists are getting played because the game is rigged in favor of speed over scrutiny. When a VC-backed startup announces it has crossed $50 million ARR, reporters face a binary choice: publish the number and win the traffic, or spend days interrogating its construction and lose the scoop to a competitor who won’t bother. Competitive pressure almost always wins.

The deeper problem is structural ignorance, not laziness. Founders and their PR teams hand journalists a single ARR figure stripped of any methodology. That number could be a single strong month multiplied by twelve. It could bundle one-time enterprise pilots that will never renew. It could ignore churn entirely. The journalist has no way of knowing, and in most cases, no one on the PR side is volunteering the breakdown.

Scott Stevenson, CEO of legal AI startup Spellbook, called this out directly on X, describing it as a “huge scam” and naming the biggest funds in the world as active participants in misleading press coverage. His post landed because it named the mechanism clearly: inflated ARR figures generate headlines, headlines validate funding rounds, and funding rounds inflate the figures further. Journalists are a critical link in that chain.

What the media ecosystem is missing is a baseline standard — a simple methodology checklist that reporters can demand before publishing any ARR figure. Does this number annualize a single month of revenue? Does it include non-recurring contracts? What is the net revenue retention rate underneath it? These are not exotic questions. They are the minimum a finance editor at any serious outlet should require. The fact that no such standard has been adopted across tech media is not an oversight — it is an opening that founders and VCs exploit every funding cycle.

Until journalists treat ARR as a derived figure that requires sourcing rather than a fact that can be quoted at face value, the inflation machine keeps running. Every unchallenged headline is a permission slip for the next one.

The real-world consequences when the numbers unravel

When ARR figures collapse back toward reality, the damage spreads far beyond the founders who inflated them.

Honest founders take the first hit. A legal AI startup reporting $4 million in genuine contracted recurring revenue looks like a laggard next to a competitor claiming $40 million in “ARR” built from annualizing a single month of API calls and pilot fees. Investors pattern-match against inflated peer benchmarks, which means companies with clean books get lower valuations, worse terms, or no term sheet at all. The inflation game punishes integrity.

Employees bear real costs too. Engineers, salespeople, and operations hires join companies based on growth narratives anchored to those headline ARR numbers. They accept below-market salaries in exchange for equity priced off valuations that assume the revenue is real and durable. When the numbers unravel — when enterprise pilots don’t convert, when usage-based revenue craters, when customers who were never truly “under contract” simply stop paying — the equity evaporates and the people who bet their careers on the story are left with nothing.

Customers and late-stage investors face the same trap. A procurement team that signs a multi-year contract with a vendor partly because that vendor looks financially stable — $50 million ARR, just raised a $200 million Series C — is making a risk assessment built on fabricated inputs. Secondary market buyers and crossover funds that purchased shares at valuations derived from inflated metrics have no clean legal recourse when the company quietly restates its numbers or simply fails to grow into them.

The historical echo is sharp. Dot-com era companies counted page views as proxies for revenue. WeWork called itself a technology company to justify a $47 billion valuation built on little more than short-term sublease arbitrage. The AI moment has its own vocabulary — “ARR,” “deployment commitments,” “pipeline value” — but the mechanism is identical: dress up a fragile or fictional number in credible-sounding terminology, get the press release out, and let the valuation mythology compound before anyone checks the receipts. The cycle ends the same way every time. The only variable is how many people get hurt before it does.

What a fix would actually require

Fixing this problem requires action from three groups simultaneously — and none of them can wait for the others to move first.

Founders with genuine traction need to stop playing the ARR inflation game entirely. The competitive move now is radical transparency: publish GAAP revenue figures, or define your ARR methodology explicitly and make it auditable. Scott Stevenson of Spellbook did exactly this when he called out the “huge scam” publicly on X, naming the pattern and refusing to participate in it. That kind of public commitment is harder to walk back than a quietly inflated metric, which is precisely why it works as a credibility signal. Startups that do this cleanly will stand apart when the correction arrives — and they will attract investors who are specifically fleeing the companies that don’t.

VCs have the most leverage and the most to lose. The biggest funds in the world, as Stevenson directly stated, are actively supporting misleading ARR figures to generate PR coverage. That is not a passive failure of oversight — it is a choice. Firms that want to avoid reputational damage when portfolio companies collapse under scrutiny need to set reporting standards internally now, before regulators or a high-profile blowup forces the conversation. Requiring portfolio companies to report to a consistent, defined ARR methodology is a board-level governance decision. It is available today.

Journalists and editors carry accountability too. Any ARR figure that gets published without a methodology attached is functionally unverified information. Public companies face mandatory disclosure standards for their earnings figures. AI startup ARR numbers get repeated in major outlets with no equivalent scrutiny. The editorial standard needs to be explicit: either the methodology behind the figure runs with the story, or the figure gets labeled as unverified. Editors who apply the same rigor to a private startup’s ARR claim as they would to a public company’s earnings call will force the entire ecosystem toward more honest disclosure — not because founders and VCs suddenly develop integrity, but because the reputational cost of misrepresentation finally becomes real.