The Justice Department expects a corporation to learn from its mistakes, as a person would. Boeing's culpability score was the same before it breached its deal and after.
31 October 2026
There is a public database that counts what companies operating in the United States have been penalized for since 2000. Bank of America sits at the top of its main table: 345 records and $89.2 billion. Sort the same hundred companies by number of cases and a different one leads, Berkshire Hathaway with 3,899, followed by Exxon Mobil with 653 and Walmart with 635. In a person, a column like that has a name, and the name changes the sentence.
The name is recidivism, and the law applies it to companies too. The Justice Department's manual for prosecutors says so in as many words: "A corporation, like a natural person, is expected to learn from its mistakes." The federal sentencing rules have a section on an organization's prior history. Prosecutors have the rule and the vocabulary. The question is what either one does.
Start with the objection, because it is correct. These numbers have no base. A firm the size of Bank of America will collect more records than a corner shop for the same reason it collects more parking tickets.
The arithmetic is rough but it points one way. Bank of America employs about 213,000 people. Its 345 records over 27 calendar years come to about thirteen a year, or one for every 16,700 years of employee time. A shop with five staff, offending at that rate, would collect its first record after about 3,300 years. Today's headcount stands in for one that moved, and big banks are watched far more closely than shops. I found no published rate that corrects for size, for any large company. The absence is worth knowing about: I cannot tell you that a firm with 345 records offends more often than one with three.
The money does not look like a rounding error either. From 2007 to 2025 Bank of America reported $1,748.5 billion of revenue and $314.3 billion of net income. The tracker's $89.2 billion is 5.1 percent of the first and 28 percent of the second. Those are imperfect ratios, because the penalty count starts in 2000 and includes class actions, and because net income is what remained after the bank paid. About seven-tenths of the penalties are mortgage and toxic-securities cases from one crisis, and many of those were inherited with Countrywide and Merrill Lynch. A critic who says this column proves a culture of lawbreaking is counting numerators.
The base rate measures how often something went wrong. It says nothing about what happens to the record when it does.
Brandon Garrett, a law professor at Duke who built a registry of corporate prosecutions with the University of Virginia, counted something that does not depend on a company's size. From 2001 to 2018, the Justice Department struck 497 deferred or non-prosecution agreements with organizations, deals in which a company pays and promises reform and is not convicted. Individuals were prosecuted alongside the company in 134 of them: 27 percent.
The figure needs one correction in the prosecutors' favor. The total was swollen in 2015 by a program for small and mid-sized Swiss banks, none of which came with individual charges. Through 2012, the share was 89 of 255, or 35 percent. Either way, most of these agreements left no person facing a charge. Across all 134 cases, 447 people were prosecuted, and 34 of them were chief executives, typically former ones.
So in most of these agreements the record does not attach to a person. Garrett's other finding is that it barely attaches to the company. Of Citibank and HSBC he wrote in 2020: "They are recidivists, but they do not receive harsher penalties despite their growing criminal records." Four years earlier he had listed UBS's sequence: a deferred prosecution agreement in 2009, non-prosecution agreements in 2011 and 2012, a guilty plea by a subsidiary in 2013, and another plea pending. When UBS breached the 2012 agreement it paid $203 million, a far smaller fine than the other banks paid in the same round of currency cases.
The rulebook explains part of this. For a person, federal sentencing has a table with two axes: how bad the offense is, and how long the record is. At offense level 20, a first offender's range is 33 to 41 months. A person with thirteen or more criminal history points gets 70 to 87. The company's version is a single line. An offense within ten years of a prior adjudication for similar misconduct adds one point to a culpability score, and within five years adds two. From the starting score of 5, two points raise the fine multipliers by 40 percent.
On a first repeat, the company's proportional increase is the larger one. One substantial prior conviction moves the person's range by about 12 percent. But the person's column keeps climbing, and the company's stops at two points, whether it has one prior or twenty.
The harder limit is in the definition. A "criminal adjudication" means "conviction by trial, plea of guilty (including an Alford plea), or plea of nolo contendere." A deferred or non-prosecution agreement is none of these. A company can sign one after another and, on the criminal side of that line, its history stays blank. Garrett put it plainly in 2016: banks "could avoid consequences for repeat criminal prosecutions since they lacked a criminal record, having settled prior cases."
Prosecutors do have discretion to look further. The manual calls repeat agreements "generally disfavored" and requires senior sign-off, and a department-wide policy from March 2026 counts earlier resolutions against a company seeking to avoid charges altogether. These are real, and they are judgment calls made in private by the side negotiating the deal. In fiscal year 2025, 110 organizations were sentenced in federal court. Of the 36 sentenced under the fine rules, two had a prior violation on the record. Across all 110, 8 percent had a history of misconduct. The median fine was $500,000.
Boeing shows the machinery at full size. Two 737 MAX crashes killed 346 people. On 7 January 2021 the company signed a deferred prosecution agreement on one count of conspiracy to defraud the United States.
The agreement contains its own arithmetic. Base score, 5. Add 2 because someone with substantial authority was involved. Subtract 2 for cooperating. Total: 5. The base fine was $243.6 million, "representing Boeing's cost-savings," the money it saved by avoiding simulator training for pilots. With the score of 5 the range ran from one to two times that figure, and Boeing paid the bottom of it. The same document records "a criminal conviction from 1989," "a criminal non-prosecution agreement from 2006," and a 2015 settlement with the Federal Aviation Administration. They add nothing to the score.
Boeing also paid $1.77 billion to airlines and $500 million to the victims' families. The fraud case rested on the conduct of two technical pilots, and the Justice Department said the misconduct was neither pervasive nor "facilitated by senior management." One of those pilots, Mark Forkner, was indicted that year and acquitted in 2022. His central defense, in the words of Judge Reed O'Connor, was "that he was a scapegoat for the broader and systematic failure of Boeing's corporate culture." The judge added: "The Government agrees with that theme."
In May 2024 prosecutors told the court that Boeing had breached the agreement, by "failing to design, implement, and enforce a compliance and ethics program." In July 2024 the parties filed a plea agreement, and it calculated the score again. Base 5, plus 2, minus 2. Total: 5. The agreement itself says the calculation is the same as in 2021. A federal rule raises the score for violating a court order or probation. A deferred prosecution agreement is neither.
The fine did move. Boeing would now pay the top of the range, $487.2 million, double the 2021 penalty and the legal maximum. That is an escalation, and it is also exactly twice what the company, by its own assessment, had saved. The range itself never changed.
O'Connor rejected that plea in December 2024, over the terms for choosing an independent monitor. In May 2025 the department offered a non-prosecution agreement instead, worth $1.1 billion: the rest of the fine, $444.5 million more for the families, and at least $455 million spent on Boeing's own compliance and safety programs. The agreement runs for two years. Boeing admits the violation, and it proposes its own compliance consultant, subject to the department's approval.
On 6 November 2025 O'Connor dismissed the case. His order reads like a dissent from his own ruling. The agreement "fails to secure the necessary accountability to ensure the safety of the flying public." The government's claim of trial risk is "unserious," because it holds "a confession from Boeing, signed by the CEO and Chief Legal Officer." And then: "The Court acknowledges that it does not have the authority to deny leave because it disagrees with the Government that dismissing the criminal information in this case is in the public interest." The families went to the Fifth Circuit, which ruled on 31 March 2026 that it lacked jurisdiction to review the dismissal, and refused rehearing in May.
Across both deals Boeing committed about $3.66 billion. That is not a trivial sum. But consider what the file holds now. The admission is there. The conviction is not, so as I read the rule, a future sentencing court would add no point under its first test for a case in which 346 people died. A prosecutor may weigh it. The table will not count it.
A fine has a number, and a firm can plan around a number. The base fine in the Boeing case was calculated from what the offense saved. That is what I mean by a price: not that the sums are small, but that they are sums.
The sum also has a blind spot. It is sized to the case in front of the court, and it is a case only because someone caught it. Using the collapse of Arthur Andersen, which forced companies to change auditors, three economists estimated that only about one fraud in three at large American firms comes to light in normal times. If that rate holds, a fine must exceed three times the gain just to break even, and Boeing's legal maximum on the fine was two times the gain. John Coffee, who has studied this for decades, gives the reason it stays low: a fine large enough to deter "would bankrupt corporations", and the cost would land on "employees, creditors, and other innocent stakeholders." A person cannot be fined out of a prison sentence. A company has no cell to serve it in.
Other sanctions work differently. On 2 February 2018 the Federal Reserve barred Wells Fargo from growing beyond its size at the end of 2017, until it "sufficiently improves its governance and controls." "We cannot tolerate pervasive and persistent misconduct at any bank," said Janet Yellen, then its chair. The restriction lasted seven years and four months, and it ended when the Federal Reserve judged the conditions met, not when a sum had been paid. It is one case, and I cannot tell you what it changed inside the bank. It does show that the law can impose a cost that is not a price.
The evidence that any of this reforms a company is thin, and some of it cuts against me. A 2023 working paper by Syrena Shirley, on public companies from 2003 to 2018, finds firms settling through these agreements less likely to be prosecuted again, and no such result for firms that pleaded guilty. It also finds them more likely to break regulations afterward. And re-prosecution is a weak measure: it falls whenever prosecutors decline to act. A review of 58 studies found regulatory policy the only single approach with a significant company-level effect, and not consistently. Nothing I have read shows that a longer record and a heavier sentence reform people either. My claim is narrower: the sentencing table keeps a running count on a person, and these agreements never enter it.
The strongest case against this essay is the Sacklers.
In 2007 a Purdue Pharma affiliate pleaded guilty to a felony for misbranding OxyContin. Before that plea, according to the Supreme Court, the owners' distributions were under 15 percent of annual revenue. After it they took as much as 70 percent, about $11 billion between 2008 and 2016, draining the company's assets by 75 percent. Purdue filed for bankruptcy in 2019 and pleaded guilty to three more felonies in 2020. In April 2026 a judge sentenced it to $5.5 billion in fines and penalties, of which the government will collect $225 million. The family offered $4.325 billion, later raised, in exchange for a court order ending all opioid claims against them, including claims from victims who refused the deal.
On 27 June 2024, in Harrington v. Purdue Pharma, the Court said no, five votes to four: "nothing in present law authorizes the Sackler discharge." The dissent warned of the consequence: "there will be no $5.5 to $6 billion settlement payment." It was wrong. The family returned to the table. The new settlement, $7.4 billion in all, took effect on 1 May 2026, with up to $6.5 billion reported from the Sacklers over fifteen years and more than $1.5 billion paid that day. It ends their control of the company and permanently bars them from selling opioids in the United States.
This is a real refusal, and it has to be weighed. The owners of a company with two rounds of felony pleas in its family were refused the protection they had bargained for, by the highest court in the country, and the deal they struck afterward is larger.
Look at what the Court did, though. It read the bankruptcy code, confined itself to the question presented, and left consensual releases alone. The prior guilty pleas did not decide the case. The ruling turned on whether a statute reaches people who did not consent, and it would have read the same for a firm with no record at all. The Justice Department charged the company and not its owners, and at Purdue's sentencing in April 2026 the judge said that was why she could not jail them. The one Sackler felony I found is Joss Sackler's guilty plea in May 2026, for obstructing a grand jury over a seized drug shipment; her lawyer calls it unrelated to Purdue. The law's firmest refusal in this story raised a price.
Return to the database. Bank of America, 345 records. Without a denominator, that count cannot say whether the bank offends more often than a corner shop. It shows a company that has been through the process hundreds of times, and the record is a list of payments.
In a person, the record is kept by the court and read out at the next hearing, and the next sentence depends on it. The Justice Department says a corporation, like a natural person, is expected to learn from its mistakes. The conviction is what would hold the lesson, and the agreement exists to avoid the conviction. Boeing's score was 5 before the breach and 5 after it.
The ideas, the argument and the conclusions in this essay are mine. I develop each essay in conversation with an AI assistant: I set the question and the direction, it turns my thoughts into prose, and I review and rework every draft until it has the shape and form I want. Every figure is sourced, and the sources are listed below.