Kenneth Roth: An Algorithm Is Deciding Who Lives and Dies in Gaza

An algorithm has been deciding who lives and who dies in Gaza. That is the central claim of Kenneth Roth, the former longtime executive director of Human Rights Watch, in a new commentary arguing that the debate over autonomous weapons has fallen behind the reality of the battlefield.

Roth's piece, published in The Guardian on October 2, 2026, takes aim at a gap in how the public conversation about AI risk is framed. When tech leaders warn about artificial intelligence — and Roth points to reporting on Anthropic's chief executive Dario Amodei and an AI slowdown — they tend to conjure images of AI agents running amok: hacking the systems that run critical infrastructure, banks, or even militaries. Those scenarios make headlines. A different danger, Roth argues, gets mentioned far less often.

That danger is AI-empowered lethal weapons — fully autonomous versions of which are colloquially known as "killer robots."

Why Gaza changed the argument

For years, discussions of autonomous weapons have proceeded on the assumption that the technology was arriving — someday. Treaties, UN expert discussions, and campaigns such as the Stop Killer Robots movement have all operated against a horizon of future deployment. Roth's argument is that Israel's conduct in Gaza has collapsed that horizon. In his words, such AI weapon systems "are already here."

The claim is not rhetorical. Roth links it to documented reporting on the Israeli military's use of digital tools in Gaza. A Human Rights Watch question-and-answer document from September 10, 2024, examined how the Israeli military has used digital tools — including AI systems — to generate targets in Gaza, and raised questions about the role these systems play in determining who is marked for death. The International Committee of the Red Cross has likewise published guidance on what autonomous weapons are and why they raise urgent humanitarian and legal concerns.

The core of Roth's warning is simple: in an important respect, the decision over life and death in a modern conflict has already been delegated, at least in part, to a machine.

Humans must remain at the center

Roth's subtitle states his prescription plainly: "Humans must remain at the center of wartime decisions." He frames Israel's conduct in Gaza as the demonstration case for why that principle matters — an example of what happens when algorithmic target generation is embedded in a real military campaign, with real casualties, rather than debated in the abstract.

The stakes are legal as well as moral. The framework that governs warfare — international humanitarian law, with its requirements of distinction and proportionality — rests on human judgment: a human assessing whether a target is military or civilian, a human weighing the expected harm against the military advantage. An algorithm does not carry responsibility. It cannot be court-martialed, and it cannot exercise mercy. Roth's argument, aligned with the position of the ICRC and the Stop Killer Robots coalition, is that this judgment cannot be outsourced.

A gap between AI-doom rhetoric and present reality

Roth's commentary also functions as a critique of the AI safety conversation itself. The loudest warnings about AI, he notes, come from tech leaders who focus on speculative runaway scenarios — systems hacking infrastructure, banks, militaries. Those scenarios are worth taking seriously, and Roth does not dismiss them. But they share a feature: they are hypothetical. The autonomous-weapons problem, by contrast, is not a thought experiment. People have already died in a conflict where, according to the human rights documentation Roth cites, an algorithm has played a role in selecting them.

That asymmetry shapes policy. Fear of future catastrophe drives attention and funding toward far-off alignment problems while a present, documented use of AI in killing receives comparatively little scrutiny. Roth's piece is an attempt to correct that balance — to insist that the most urgent AI risk discussion should include the weapons systems already operating.

Why this matters now

The debate over autonomous weapons has been running for more than a decade at the diplomatic level, with campaigns pressing for a binding treaty prohibiting systems that select and engage targets without meaningful human control. Governments have so far failed to agree on one. Every real-world deployment of AI targeting — and Gaza is now the most extensively documented case — becomes evidence in that argument, either for those who say existing law suffices or for those, like Roth, who say the technology has outpaced it.

Roth writes from a position of institutional authority on these questions. As executive director of Human Rights Watch from 1993 to 2022, he spent nearly three decades documenting wartime abuses and pressing for legal accountability. His decision to frame Gaza as proof that AI weapons have arrived signals how human rights organizations now read the conflict: not only as a humanitarian catastrophe but as a precedent for how machines will be used in war going forward.

The question his piece leaves the reader with is the one the treaty negotiations have never resolved: if an algorithm is already deciding who lives and who dies, what legal barrier will stop the next military — and the one after that — from letting it decide more? Until governments answer that, the strongest constraint on autonomous killing remains the one Roth invokes: the insistence that a human being must stay in the loop.