Gonzalez v. Google: When the Algorithm Itself Is the Evidence

Picture nine of the most powerful judges in the United States, each one used to having the last word, suddenly out of their depth, not on a point of law but on a point of machinery (cue nine black robes squinting at what amounts to a flowchart). For an entire afternoon, a courtroom full of lawyers and Supreme Court Justices tried to put into plain words what a recommendation algorithm actually does. Not whether it’s good or bad, or whether it should be regulated. Just what it is, mechanically, and where “showing you a video” stops being hosting and starts being something else. That it tripped up the nine most senior judges in the United States says less about American free-speech politics than about what happens in any courtroom, anywhere, including Israel, when a dispute turns on software nobody in the room can describe with precision.

What actually happened

It starts with a loss no courtroom can fully hold. Nohemi Gonzalez, a 23-year-old exchange student from California State University, Long Beach, was killed in the November 2015 Paris terrorist attacks, which ISIS claimed. Her family did something that, at the time, had never quite been tried this way: they sued Google in the United States, arguing that YouTube’s recommendation algorithm actively pushed ISIS recruitment and propaganda videos to users, and that this went beyond merely hosting third-party content. The family brought the claim under the U.S. Anti-Terrorism Act, arguing that Section 230 of the Communications Decency Act, the law that shields platforms from liability for content posted by others, should not extend to a platform’s own algorithmic choice to recommend that content. Google’s position was the mirror image: that recommending third-party content is just another way of organizing and displaying it, and Section 230 protects that the same way it protects hosting it in the first place.

The Supreme Court heard oral argument on February 21, 2023. What came out of that argument wasn’t a ruling on the merits of either side’s algorithm theory. It was something narrower, and for anyone who works with technical evidence, far more revealing. Justice Alito told the family’s attorney he was “completely confused by whatever argument you’re making,” Justice Jackson said she was “thoroughly confused” about how the case was supposed to work, and Justice Kagan put it more bluntly still: the Court, she said, doesn’t “really know about these things.” These are not, in her words, “the nine greatest experts on the internet” (an honest admission from people whose day job is drawing clean lines). Then, months later, came the moment everyone was waiting for, and it wasn’t the answer anyone expected. In May 2023, the Court issued its decision: it vacated and remanded the Ninth Circuit’s judgment, not because it resolved the Section 230 question, but in light of its own ruling that same day in the companion case Twitter v. Taamneh, which held that the plaintiffs in that related case had failed to state a viable claim under the Anti-Terrorism Act regardless of how Section 230 applied. The Court explicitly left the algorithmic-recommendation question under Section 230 unresolved, siding with neither Google nor the Gonzalez family. It punted (the legal equivalent of finishing a marathon and declining to say who won).

The artifact: a recommendation system nobody could fully characterize in court

Nothing here was hacked, stolen, or leaked. There’s no breach log, no exfiltrated database, and that’s exactly what makes this case so unsettling for anyone who thinks of “evidence” as something you can point to. The disputed artifact is the recommendation algorithm itself: a system that ingests a user’s watch history and engagement signals and outputs a ranked list of “next” videos, at a scale and speed no single person could review by hand. The entire legal fight hinged on how that system should be characterized: is ranking and surfacing content an editorial-adjacent, essentially passive extension of hosting, or is it an independent, system-generated act that creates its own liability? Lawyers on both sides needed to reason about this system’s behavior in front of a bench that, by its own members’ admission, was not equipped to open it up and describe its real inputs and outputs from the inside. Nine of the country’s sharpest legal minds ended up describing a black box from the outside, like reviewing a restaurant off the menu seen through a frosted window. That gap, between the legal question and the technical reality underneath it, is exactly where a case like this stalls.

What a court software/algorithm expert witness actually tests in a recommender-system case

This is the gap a computer forensics expert witness exists to close. A recommendation or ranking system is not a black box that a lawyer can describe from a press release or a plaintiff’s intuition about “the algorithm pushed extremist content.” It is a piece of software with real inputs, real logic, and real outputs, and those can be examined. In a recommender-system dispute, a court-appointed technical expert doesn’t argue about “algorithms” as a political abstraction. The expert narrows the question to something testable: what specific signals fed this particular recommendation, what ranking logic (or model behavior) produced it, and how is that different from, or the same as, a platform simply hosting a page a user could reach directly? From there, the expert’s job is to produce a written opinion that gives the court a precise, falsifiable account of the system’s actual behavior: not a metaphor, not an analogy to a newspaper editor or a library shelf, but a tested account of the code and data path. That is the record a court needs before it can reason clearly about where “hosting” ends and “recommending” begins, instead of leaving the bench to do what happened in open court in February 2023: guess at the line.

Why this matters for Israeli litigation

Here’s the part that should keep an Israeli litigator up at night: the Supreme Court had every resource in the world and still walked away without an answer. Israeli courts are going to see more of exactly this problem, not less. Any dispute involving a recommendation engine, a ranking algorithm, a matching system, or an AI model’s output is going to turn on the same underlying question Gonzalez never answered: what did the system actually do, and can anyone describe it precisely enough for a judge to rule on it? A litigator who walks into that kind of case with only an argument about what the algorithm “probably” does, instead of a tested technical account of what it provably does, is in the same position the Justices were in during oral argument, reasoning about a system from the outside. Bringing in a computer forensics expert witness early, before positions harden, is what turns “the algorithm did something bad” into a claim a court can actually evaluate.

The above is general information and not legal advice. Case facts are drawn from the sources listed.