Computer Associates International, Inc. v. Altai, Inc. is a 1992 decision of the United States Court of Appeals for the Second Circuit on how far copyright protects the “non-literal” elements of a computer program, meaning the aspects not reduced to written code, such as structure and organisation. The court adopted a three-step “abstraction-filtration-comparison” test and held that Altai’s rewritten OSCAR 3.5 program did not infringe Computer Associates’ (CA) CA-SCHEDULER program. An earlier version of Altai’s program, OSCAR 3.4, contained code copied by a former CA programmer; Altai did not appeal the USD 364,444 awarded to CA for it.[1] Courts have continued to apply the test to software; the Federal Circuit cited it in SAS Institute v. World Programming in 2023.[2]
CA’s software business later became part of Broadcom; see Broadcom mainframe software licensing.
Background
CA-SCHEDULER was a job scheduling program for IBM mainframes. It contained a component called ADAPTER, which the court described as an “operating system compatibility component”: it translated the program’s requests into calls that the installed operating system (DOS/VSE, MVS or CMS) could understand, so the task-specific part of the program did not need to be customised for each operating system. ADAPTER was not sold separately.[1]
Altai marketed its own scheduling program, ZEKE, from 1982. In 1983 it decided to make ZEKE run on MVS. Its employee James Williams recruited Claude Arney, a CA programmer who knew ADAPTER. When Arney left CA in January 1984, he took copies of the source code for the VSE and MVS versions of ADAPTER, in knowing violation of the CA employee agreements he had signed. At Altai he proposed a common system interface component, called OSCAR, without telling Williams that the idea came from ADAPTER, and he wrote OSCAR using the ADAPTER source code. About 30 percent of OSCAR’s code was copied from ADAPTER. The district court accepted that no one else at Altai knew.[1]
The dispute
CA learned in July 1988 that Altai might have appropriated parts of ADAPTER, registered copyrights in versions of CA-SCHEDULER and sued for copyright infringement and trade secret misappropriation.[1]
On receiving the complaint, Altai learned from Arney which code he had copied. On advice of counsel, Williams started a rewrite. Arney was excluded and his copy of the ADAPTER code was locked away. Eight programmers who had not worked on OSCAR 3.4 were given a description of ZEKE’s operating system services and rewrote the affected code over about six months, finishing in November 1989. Altai then shipped only the resulting OSCAR 3.5 and gave it to existing customers as a free upgrade.[1]
After a six-day trial in 1990, the district court found that OSCAR 3.4 infringed and awarded CA USD 364,444 in actual damages and apportioned profits. It found that OSCAR 3.5 was not substantially similar to ADAPTER, and that CA’s trade secret claim was preempted by the Copyright Act. CA appealed the last two rulings. Altai abandoned its own appeal and conceded liability for OSCAR 3.4.[1] CA argued that the district court’s test did not account sufficiently for a program’s non-literal elements.[1]
Decision or outcome
The Second Circuit (Judges Altimari, Mahoney and Walker; opinion by Judge Walker) affirmed the copyright ruling and vacated and remanded the trade secret preemption ruling.[1] It found the results of earlier software cases “less than satisfactory” and set out a three-step test for substantial similarity of non-literal program elements:[1]
- Abstraction. Drawing on Learned Hand’s abstractions test in Nichols v. Universal Pictures, the court dissects the allegedly infringed program into its structural parts at levels of increasing generality, in a manner the court likened to reverse engineering on a theoretical plane.[1]
- Filtration. At each level the court filters out material that is an idea, that is dictated by efficiency, that is required by factors external to the program, or that is taken from the public domain. The court listed external factors such as the mechanical specifications of the computer, compatibility requirements of other programs, manufacturers’ design standards, the demands of the industry served, and widely accepted programming practices. It applied the doctrines of merger and scènes à faire to software.[1]
- Comparison. What remains, the “core of protectable expression”, is compared with the defendant’s program, considering both whether it was copied and how important the copied part is to the plaintiff’s program as a whole.[1]
Applying this approach, the court agreed with the district court that OSCAR 3.5 did not copy protectable expression from ADAPTER.[1] It acknowledged that it may seem counter-intuitive for a party that had benefited from illicitly obtained material to emerge “relatively unscathed”, but said that where the appropriated material is non-protectable expression this is how copyright advances the progress of science and art.[1]
On trade secrets, the court held that CA’s claim was not preempted to the extent it rested on breach of a duty of confidentiality, and remanded it. The court said that trade secret law, used without encroaching on the Copyright Act, is an appropriate means to secure compensation for “software espionage”. Judge Altimari concurred in the copyright part and dissented from the trade secret part of the amended opinion, saying he adhered to the panel’s original opinion of June 1992.[1]
Significance for software licensing and SAM practice
Altai is a leading US authority on the scope of copyright in software beyond its literal code. Its filtration step removes from protection elements dictated by compatibility and efficiency, which bears directly on products built to work with, or replace, another vendor’s software.[1] In 2023 the Federal Circuit noted that the Second, Fifth and Tenth Circuits, among others, have adopted the test to determine the scope of protection for computer programs, including their non-literal elements, and affirmed a finding that SAS had not shown its asserted elements were protectable.[2]
The case also shows the limits of copyright as a tool against a competitor’s compatible product, and the role that contracts and confidentiality obligations play alongside it. Later disputes over mainframe compatibility and emulation, such as IBM v. LzLabs, have turned on licence terms rather than on copyright alone.
Lessons learned
- Separate function from expression. The court filtered out ideas, efficiency-driven design, compatibility requirements and public domain material before comparing programs.[1] When a vendor claims that a compatible or replacement product infringes, the question under US law is whether protectable expression that survives filtration was copied.
- Clean-room rewrites can cure literal copying. Altai excluded the developer who had copied code, locked away the copied source, and had other programmers rewrite from a functional description. The resulting program was held not to infringe, although Altai remained liable for the earlier version.[1] Document such processes if you need to rely on them.
- Confidentiality obligations still matter. The copying began with an employee who took source code in breach of his employee agreements, and the court kept CA’s trade secret claim alive to the extent it rested on a duty of confidentiality.[1] Control who can take source code off premises, and screen new hires for material from former employers.