That old chestnut showed up again in tonight's digest — the one I've honestly never fact-checked: supposedly, under the 1956 antitrust consent decree, AT&T couldn't sell software, so they gave Unix to universities for almost nothing, and the entire server world grew out of that regulatory scar. Beautiful story. And it's almost true — which makes it dangerous.
I went down to the primary sources: the newspaper front page from signing day, the actual court documents, memoirs from the guy who sold those licenses, and the 1993 ruling where Unix's copyright effectively disintegrated before a judge's eyes. Here's what I found: the 1956 decree said absolutely nothing about software. No ban existed — what existed was a feeling that it was better not to poke the judicial dragon. And that's where it gets interesting.
It started, absurdly enough, with hardware. In January 1949, the Justice Department sued AT&T for monopolization: the telephone monopoly owned Western Electric, the sole supplier of all Bell System equipment, and was buying from itself at inflated prices. The government wanted a hard solution — strip away Western Electric and split up its assets. This was the heyday of classic antitrust: RCA and IBM had gotten similar treatment in the preceding years.
Then seven years of nearly empty files. But behind the scenes, things were boiling, and in 1959 a House subcommittee cracked that backstage world open. February 1952: Bell Labs president Mervin Kelly visits Defense Secretary Robert Lovett asking to suspend the case until the end of the Korean War — AT&T, you see, is a critical defense contractor (it ran Sandia, the nuclear arsenal). The Attorney General refused. After Eisenhower's team came in, AT&T pressed again, and on June 27, 1953, at a judicial conference in White Sulphur Springs, company counsel Thomas Price met with the new Attorney General Herbert Brownell. According to Price's own memo, the Attorney General said they "needed to find a way to get rid of the case," and that AT&T would "easily find practices it was willing to have enjoined without real damage to the business." This isn't conspiracy theorist paranoia — it's a quote from a document later reproduced in the 1983 court decision.
The cherry on top: Defense Secretary Charles Wilson sends the Attorney General a letter asking him not to touch Western Electric because splitting it up "would be contrary to the vital interests of the nation." Judge Harold Greene noted thirty years later with dry irony: Wilson's letter had actually been drafted by AT&T. The lobbying machine worked at all levels simultaneously, and by December 1955 the deal was done.
On January 24, 1956, federal judge Thomas Meaney in Newark signed the Consent Decree — a settlement agreement that took effect as a court order, without a single hour of trial. The next day, the New York Times ran a front-page headline: "A.T. & T. SETTLES ANTITRUST CASE; SHARES PATENTS." Government lawyers called the deal "one of the most important in antitrust history," one of them called it "wonderful." AT&T president Cleo Craig publicly grimaced: the terms were "stringent."
What was actually in there:
Among those 8,600 liberated patents were patents on the transistor — that same year its inventors would get the Nobel. This is one of the twentieth century's most underrated gifts: an entire Silicon Valley grew from a document written to punish a phone company.
Now the main detail, the reason I went into the archive. The word "software" does not appear in the decree text. At all. Not one clause, not one provision. In 1956 there was no software market — programs came bundled with machines, like instructions for a toaster. The ban was formulated as "business other than common carrier communications services," and the lawyers of 1956 were thinking about telephone sets and power plants, not operating systems that didn't yet exist.
Otto Wilson — the man at AT&T who designed the actual Unix licensing structure — would write about this in 1985 with all the directness of an insider: "Contrary to popular opinion, the Decree did not deal with software. The technology was virtually unknown at that time. Thus AT&T was neither obligated to license its software nor restricted from entering into the software business. Software was a muddy issue in an otherwise clear statement." Remember that phrase — it breaks half the popular retellings.
What was happening at Bell Labs at this time is familiar to anyone who's read about Unix, so just the highlights. 1969: Multics failed, Ken Thompson finds an idle PDP-7 and writes a system for himself that Brian Kernighan would later dub Unix in a pun on Multics. The official request for a proper machine (PDP-10) fails — Research VP Bill Baker supposedly summed it up: "Bell Laboratories just doesn't do business this way!" 1971: the team gets a PDP-11 under the pretext of a "text processing system for the patent department" — which, ironically, became Unix's first real user, because it was patent lawyers who handled the licensing obligations from the decree. 1973: kernel rewritten in C. 1974: paper in CACM — and a flood of requests.
And here the legal machinery of 1956 unexpectedly collides with technology. AT&T's gentleman's formula, which Wilson repeated "at every gathering of believers," went like this: "As is, no support, payment in advance." The company's own retrospective report formulates the policy in four points: no advertising, no support, no bug fixes, payment in advance.
Peter Salus noted in BYTE in 1994 a consequence that's hard to overstate: the refusal to provide support "had an immediate effect — it forced users to share with one another." Ideas, programs, bug fixes, hardware repairs. May 15, 1974, twenty people gathered in a Columbia University conference room — the first Unix User Group meeting, future USENIX. July 1975: Mel Ferentz from Brooklyn College publishes the first "Unix news" report, circulation — 33 institutions. September 1976 — already 138, 37 of them outside the US. In 1977, Interactive Systems of Santa Monica becomes the first company to sell commercial Unix support: the support market was born outside, AT&T itself had nothing to do with it.
Berkeley — Ken Thompson returns to his alma mater as a visiting professor in 1975, Bill Joy writes vi and a Pascal compiler — releases the first BSD in March 1978: license on one side of a page, tape for $50, cost of media. Compare that to Western Electric's multi-volume confidential agreements where every employee had to confirm in writing that they'd been warned about confidentiality. Salus described the distribution cycle in one sentence: something gets created at Bell Labs, distributed in source, a user in Britain improves it, a user in California improves both, redistributes at cost — and the improvement goes into the next official release. "There was no going back."
But behind this mutual aid circle wasn't an idyll at all. Bell Labs lawyers chose trade secret regime for Unix instead of copyright — and that decision, made out of caution, would kill AT&T's Unix copyright twenty years later. Remember this turn, we'll come back to it.
Speaking of "couldn't sell": Wilson describes the internal mechanics like this. AT&T licensed software under decree procedures, "but made it explicit that it did not intend to be in the software business." Nobody wanted to bring the question up for legal review — because the answer might require renegotiating the settlement with Justice, and nobody in the company dared raise a question requiring renegotiation. This isn't a ban. This is institutional cowardice elevated to business strategy. The myth "the judge banned software sales" is a myth AT&T told itself and the market to avoid dealing with uncomfortable questions.
While Unix grew, the Bell System cash machine worked its rhythm. The decree obligated Bell Labs to license everything funded by the regulated business — for this they created Patent Licensing, an office with lawyers, accountants, and million-dollar deals: 4–5 serious negotiations a year. Software licenses against that backdrop — dust: by the early 70s AT&T had issued a couple dozen software licenses, ten of them promising, the leader — Unix.
Then the avalanche. Universities write themselves (1967 — before Unix even, simple letter agreements, royalty-free through patent lawyers). 1974: they introduce nominal fees for universities — revenue doesn't even cover media and shipping costs. First commercial requests stump the company: there's no price. Wilson honestly describes pricing debates: (a) price high to discourage; (b) price low to encourage experiments; (c) market price minus adjustment for no support. They chose market — and thereby silently acknowledged that Unix has a "customer base," one step away from a "business."
Commercial clients wanted sublicenses, features, support. AT&T answered: "we have one customer — Bell System itself." The industry's love-hate relationship with Unix was born from exactly this phrase. And in 1979 and 1980 Microsoft and SCO release Xenix, Berkeley — 4BSD: the ecosystem already lives its own life, AT&T's heirs trade an inheritance the official mother refused to acknowledge.
By the end of 1980, Western Electric "buys out" the other internal financial partners in Unix, distribution moves from Bell Labs to WE — the system starts thinking like a product. Under pressure from the biggest external client, the federal government, documentation and support are born — that's how UNIX System III is born, the first release issued externally simultaneously with the internal one. The license backlog — 200+ requests — becomes unmanageable.
And here Judge Greene returns to the stage.
The second AT&T antitrust suit was filed November 20, 1974 — "only" 18 years after the first, when the FCC had already cracked open the market (Carterphone 1968, MCI 1969), and the antitrust division's reputation still smelled like Watergate. The process dragged eight years, and in August 1982 the parties agreed on the Modified Final Judgment: the 1956 decree "vacated in its entirety and replaced" — canceled entirely and replaced with structural separation. AT&T gives up 24 local phone companies into 7 regional RBOCs, but gets freedom to enter any market. Greene writes directly in the 1983 decision: the new decree will free AT&T for "the computer market and the full range of information services," and there were no serious objections to the company entering computer manufacturing.
In January 1984 the divestiture happened. Ritchie noted in that year's Turing lecture with a researcher's weary irony: "AT&T's entry into the computer industry will test, and perhaps strain, this balance." Wilson is chronologically precise: in January 1983, weeks after the MFJ announcement, AT&T announced UNIX System V — with training and tech support. The formal entry into the software business happened the moment the ban — or rather, the fear of it — disappeared. The 200+ request backlog cleared in six months, turnaround dropped from 4 weeks to 1.
Now the tragic irony that holds this entire story together. That same legal caution that grew from the decree, in 1975 forced them to remove copyright notices from Unix source code — they feared notices would undermine trade secret status. In 1979 the research version 32V was distributed to thousands of licensees without notices. Under copyright law in effect until 1989, distribution without mandatory notice meant general publication and loss of copyright — forever.
In 1992 this backfired. The subsidiary Unix System Laboratories sued Berkeley Software Design and the University of California: trademark, false advertising, then — copyright and trade secrets. Judge Debevoise (District of New Jersey) in the ruling of March 3, 1993, dissected the claim methodically, like a surgeon. Could "carefully selected corporate clients" be considered a "selected group" under the limited publication doctrine? No: "USL wants what any landlord wants — someone who will take care of the property and pay their bills. If that were enough, virtually any lessee would belong to a select group." And the verdict: "USL has not demonstrated a likelihood of successfully defending its copyright in 32V." Injunction denied.
Simultaneously Berkeley filed a counterclaim in California court: System V, supposedly, is itself stuffed with Berkeley code without the attribution required by the license USL itself signed. Mutually assured destruction: if USL wins and raises the standard of proof, that same standard will kill it in California. In February 1993 Novell bought USL from AT&T — and inherited a case that couldn't be defended. The February 1994 settlement reads like capitulation: out of 18,000 Berkeley files, three to delete and ~70 — add USL notice. 4.4BSD-Lite declared "clean." Novell bought itself a quitclaim and buried the existential question about copyright under a sealed settlement.
Unix's legal fate: the copyright they sold, transferred from Novell to Santa Cruz Operation, then to Caldera/SCO, — it probably never existed. Everything that grew from BSD — FreeBSD, OpenBSD, NetBSD, macOS ultimately, — grew from code whose owner lost the lawsuit for the right to own it.
Let's assemble the whole line, because it's more beautiful than any fiction:
This entire arc — from prohibitive monopoly to open source — passed through one document in which the word "software" appears not once.
First. The popular formula "the decree banned AT&T from selling software" is historically inaccurate. It's accurate in its function: something was banned, but the subject was businesses unrelated to telecom services. Software in 1956 was simply invisible to lawyers. The real mechanism wasn't the letter of the decree, but its interpretation: the extreme caution of a corporation that didn't risk finding out where exactly the boundary ran. Wilson called it "muddy issue in an otherwise clear statement." The judicial dragon from Justice didn't ban selling Unix — it was just too scary to ask its opinion.
Second. This is an engineering lesson about implicit constraints. A system's real architecture is determined not by what's explicitly prohibited, but by what you're afraid to clarify. AT&T had the formal right to sell software — and for 25 years lacked the moral right to find that out. The entire history of open software grew from a misunderstood document: one inaccurate interpretation gave the world USENIX, BSD, and Linux. Worth remembering when someone's project has "for obvious reasons" that prohibit doing X: very often X is prohibited not because it can't be done, but because someone once was afraid to ask.
Third. The final irony, without which the story is incomplete. When the decree disappeared (1982–1984), AT&T immediately turned Unix into a product — and exactly this commercialization, together with USL's final lawsuits, killed proprietary Unix. The freedom to sell turned out worse than the ban on giving away. The system they couldn't sell conquered the world through university tapes at $50 and "as is" — while the system they finally sold dissolved in trademark wars. The 1956 decree, knowing nothing about software, turned out to be the best product manager in industry history — the one that for twenty-five years didn't let the phone company ruin what didn't belong to it.