Germany · de.indymedia · · 2h
Interview about AI and the Lavender System: Automated Death Lists
Deutsch (original) · Auto-translated to English
Symbolbild · Ansgar Koreng / CC BY-SA 3.0 de · Wikimedia Commons
Computer systems that automatically evaluate large amounts of data in real time are changing military target selection. An analysis of the Lavender system shows that such software violates obligations under international law. We talk to Taylor Kate Woodcock and Rainer Rehak about the consequences of “AI” in armed conflicts. They demand the immediate shutdown of such systems.
The software is crucial in attack planning and identifies potential targets for bombs in the Gaza Strip or Lebanon: Thishttps://apnews.com/article/israel-palestinians-ai-technology-737bc17af7b...“Lavender”According to media research and a recent documentary, it is intended to locate suspected members of Hamas and the Islamic Jihad of Palestine. It is intended to detect people, vehicles and other objects in real time. For the Israeli army, the software significantly accelerated and automated target selection.
In the disturbinghttps://www.theguardian.com/world/2026/sep/10/secret-systems-used-by-isr...“Naza”Soldiers involved describe the previously secret methods used by the Israeli army. The film by Juval Abraham and Rachel Szor, which was released athttps://www.labiennale.org/en/news/official-awards-83rd-venice-internati...Venice Film Festivalreceived the jury's special prize at the weekend, examines the Israeli military actions and sheds light on alleged war crimes against the Palestinian civilian population.
Software like Lavender is trained using known data sets from people in order to recognize behavioral patterns. When she then discovers new people in the field who have similar behavioral patterns, they become targets. This way I have Lavenderhttps://www.theguardian.com/world/2024/apr/03/israel-gaza-ai-database-ha...People markedwho were classified as military opponents. The computer system was used in conjunction with another software-based decision support system called “The Gospel,” which suggested buildings and structures as targets.
The civil damage and human suffering that such a system causes raises ethical and legal questions that science is also addressing. We talk to Taylor Kate Woodcock and Rainer Rehak about theirshttps://doi.org/10.1145/3805689.3812357">Lavender analysis“Automating civilian harm: On Israel’s use of the AI-enabled targeting system Lavender in Gaza and International Humanitarian Law”, which they released in June of this year. They explain why the use of software systematically violates international law.
https://www.asser.nl/who-we-are/taylor-kate-woodcock-llm/">TaylorKate Woodcockresearches in the field of international law at the Asser Institute at the University of Amsterdam.https://www.weizenbaum-institut.de/en/portrait/p/rainer-rehak/">RainerRehakis a computer scientist and researches at the Weizenbaum Institute in Berlin.
This interview is alsohttps://netzpolitik.org/2026/interview-on-lavender-civilian-harm-is-pre-...in English.
Lavender as part of the “kill chain”
netzpolitik.org: You havehttps://doi.org/10.1145/3805689.3812357">thatLavender target system and its consequences for international lawanalyzed. Please briefly describe what the Lavender system is and what it is intended to achieve for the Israeli military.
Taylor Kate Woodcock: The Israeli military describes Lavender as a sophisticated system for tracking and targeting militants in the Gaza Strip. It is a data processing system supported by “artificial intelligence” that is used to calculate attack targets.
So it is not an attack system, but part of an early phase of what experts call a “kill chain”. The results of Lavender flow into coordination and scheduling systems and ultimately end up in the attack plans of fighter aircraft, bombers, artillery, drones or ground troops, which then carry out the actual attacks.
Lavender creates a risk score between 1 and 100 for each person in Gaza, which is intended to indicate the likelihood that that person is a member of or associated with a militant group, such as the military wing of Hamas (Qassam Brigades) or the Islamic Jihad of Palestine. Since every person in Gaza receives such a value, a threshold can be set in the system. Anyone who exceeds this value becomes a military target.
netzpolitik.org: How exactly does this review of Lavender work?
Rainer Rehak: The occupation's control of Gaza allows Israel to collect vast amounts of data on virtually everything within the Gaza Strip, from satellite imagery, cell phone location data, video surveillance footage, geospatial data, communications and metadata, building information, movement data, payment data, drone footage, and even information about family relationships and social structures, as we have seenhttps://www.972mag.com/lavender-ai-israeli-army-gaza/">unabhcurrent journalistic projectsfrom the United States and Israel know.
All this heterogeneous data is linked and combined to serve as the data basis for a method called Positive Unlabeled Learning (PUL). So an “AI” system uses the data from a few confirmed cases of military actions to statistically analyze the rest of the data and create some kind of statistical similarity between all the data compared to the known cases. This statistical “similarity” is then converted into the score that indicates the risk value.
“Around 1,000 potential targets every day”
netzpolitik.org: This is also what your analysis is abouthttps://www.972mag.com/mass-assassination-factory-israel-calculated-bomb...second “AI” system, “The Gospel”. What is “The Gospel” used for?
Rainer Rehak: There is much less information about the Gospel System, but generally speaking it is a system for classifying the buildings in Gaza according to their purpose, from civilian private or commercial use to military use. The aforementioned data will be used accordingly.
netzpolitik.org: What sources and data bases do you base your analysis on?
Taylor Kate Woodcock: We rely on multiple sources of investigative reporting, using anonymous sources within the Israel Defense Forces (IDF) and U.S. contractors, as well as confidential documents. In particular, we relied on the analyzes ofhttps://www.972mag.com/lavender-ai-israeli-army-gaza/">Yuval https://www.972mag.com/mass-assassination-factory-israel-calculated-bomb...as well as the work of Michal Biesecker, Sam Mednick and Garace Burke, who worked as journalists at thehttps://apnews.com/article/israel-palestinians-ai-technology-737bc17af7b...Pressresearch.
A challenge in studying the use of AI in military operations is that how these systems are actually used in practice and what results are achieved are shrouded in secrecy. However, we have found that the reports in these independent media are supported by the technical plausibility of the systems described and also by the fact that they are based on anonymous sources with practical experience. The research explains the terrible images of destruction in Gaza that appear again and again in the media.
netzpolitik.org: Gospel and Lavender produce high target numbers. What does that mean practically? For example, how many targets does the system find per day?
Taylor Kate Woodcock: According to current estimates, around 1,000 potential targets were identified every day in the first two years of the armed conflicts in Gaza and Lebanon. According to the defense company Elbit Systems, a total of 850,000 targets were recorded in real time between October 7, 2023 and the end of 2025. These numbers suggest that the IDFhttps://www.theguardian.com/world/2026/jul/06/israel-command-system-iden...to or more than half of the entire population and infrastructure in the Gaza Striphad targeted.
Rainer Rehak: These daily figures fluctuate as the Israeli military also sets the threshold depending on available attack assets, such as bombers, and ammunition. On days when all initial targets had already been hit but fighter jets were still standing by,https://www.972mag.com/lavender-ai-israeli-army-gaza/">wasthe threshold lowereduntil the desired number of new goals was reached.
In other words, the system was repeatedly reconfigured depending on the available military resources. From the perspective of international law, this shows that the targets that were calculated never had a material basis to justify the attacks. The people of Gaza were essentially put on a big death list and then attacked from top to bottom with whatever was available.
https://netzpolitik.org/2024/40-jahre-fiff-gegen-die-unheilvolle-verflec...the disastrous intertwining of computer science and the military
Does another person decide?
netzpolitik.org: How is it guaranteed that a human being decides?
Rainer Rehak: According to the reports, each target is evaluated by a human overseer, but they only had a maximum of 20 seconds per target. He often just checked to see if it was a man. Here it becomes clear what the whistleblowers meant by the statement that thehttps://www.972mag.com/mass-assassination-factory-israel-calculated-bomb...on quantity and not on quality, which is remarkable given the fact that this is a matter of life and death.
Taylor Kate Woodcock: One danger of using AI systems is that human users place too much trust in their output and do not question it critically. This phenomenon is called “automation bias”. The risks posed by this: Even if human oversight is included in the AI systems' target selection - and in extremely short time frames - the results can still occur without real examination and verification of the decisions for or against target selection.
The extent of bulk target selection, the speed at which these AI systems create targets, and the risk of “automation bias” are fundamentally changing military target selection by reducing the time and opportunities for human review of targets. However, target selection is subject to obligations under international law, which require an examination of the legality of the targets and measures to avoid civilian casualties. If you look at the time period for each target review, this obligation under international law was clearly not fulfilled, as confirmed by the whistleblowers who described themselves as mere “stamps of approval”.
netzpolitik.org: How are civilians treated in Lavender?
Rainer Rehak: We must note that the statistical approach does not allow any serious distinction between civilians and legitimate targets because it is neither clear how exactly the score is calculated nor how the threshold is set. An internal test showed that the system itself had an error rate of about ten percent.
Additionally, the system was explicitly configured to allow civilian casualties, sometimes callously referred to as collateral damage. For those classified as unimportant, who had a lower score, the acceptable number of civilian casualties was set at 15 to 20; for high-ranking individuals the number was said to have been “hundreds.” For example, for 100 targets, the system was explicitly configured so that between 1,500 and 10,000 civilian victims would also be killed, which is probably a lower limit given the high population density in the Gaza Strip, which is comparable to that of Manhattan, New York.
The timing was often chosen so that targets were attacked not during the day but at night when they were home with their families, resulting in even more civilian casualties. This particular scheduling system is called “Where’s daddy?” (German: Where is Papa?).
Taylor Kate Woodcock: We have analyzed this practice in detail and have come to the conclusion that it systematically and seriously violates international law.
Who provides the infrastructure?
netzpolitik.org: What does the underlying technical infrastructure look like and who provides it?
Rainer Rehak: The entire Gaza datasethttps://www.newsday.com/news/nation/israel-palestinians-ai-weapons-q3224...an estimated 13.6 petabytes, i.e. the equivalent of 13,600 terabytes. This amount of data is not easy to handle, especially if you want to use it to train statistical models or even large language models that are also being created. The digital infrastructure is provided by Google and Amazon under the project name Nimbus and costs $1.2 billion.
Palantir does the integration of heterogeneous data sources. Microsoft, Cisco, Dell and Red Hat/IBMhttps://apnews.com/article/israel-palestinians-ai-technology-737bc17af7b...additional IT servicesaround the “kill chain”. This demonstrates the enormous role that the private sector plays in integrating and using AI systems in armed conflict.
netzpolitik.org: You analyze that the Israeli military systems described cause massive civilian damage and represent systematic violations of international law. What does “systematic” mean here?
Taylor Kate Woodcock: In international law there are three principles for lawful behavior in armed conflict:
- A clear distinction is made between lawful targets (such as fighters, militants or military objects and buildings) and civilians; only attacks on the former are lawful.
- Constant consideration and adherence to precautionary measures to avoid or minimize harm to civilians.
- Avoid attacks in which the expected harm to civilians is disproportionate to the expected military advantage of the attack.
When analyzing the system, it turned out that civilian damage is not an exception in isolated cases caused by errors, but rather the entire design, setup and configuration of the system inevitably leads to the large amount of civilian damage to people and infrastructure that we then read in the media and which is also admitted by the military themselves. In summary, it can be said that civilian damage is inevitable in the system and could even be considered a core function.
Such a system violates all three principles in a serious and systematic manner. Targeting systems designed in this way demonstrate a reckless disregard for the duties of selecting legitimate targets and protecting civilians, and suggest that this suffering is automated and arbitrarily inflicted.
So we cannot say that the AI system works “erroneously”, but rather that the amount of damage can actually be predicted through the deliberate configuration and use of these systems. This repeated application to thousands of targets during the conflict, as well as the deliberate decisions regarding system configuration and application, suggest that this harm is systematic and that the target selection processes do not structurally adequately ensure the safeguards required by international law.
netzpolitik.org: If the violations of international humanitarian law are inevitable and do not represent a side effect but rather a core function of the system, is the lawful use of such a system then impossible? Or could it be customized?
Taylor Kate Woodcock: Lavender and the entire “kill chain”https://www.972mag.com/mass-assassination-factory-israel-calculated-bomb...for “mass killings” To be designed with bombs in densely populated areas. Taking civilian casualties and international law seriously, the system cannot be used lawfully in its current form, including the statistical assessment, the arbitrary threshold, the predefined civilian damages and other features described in detail in our article.
Hypothetically, if we were thinking about customizing the system, we could, for example, request that the civil damage configuration be set to a value of 0, but then the system would probably not provide valid targets since Gaza is an urban area. We believe that adapting this system to the law is practically impossible.
Rainer Rehak: Against this background, it is particularly disturbing that Lavender appears to have been developed in collaboration with Israeli universities, like Maya Wind from the University of Californiahttps://www.mayaywind.com/book">inher book “Towers of Ivory and Steel”has described.
netzpolitik.org: Are there other armed conflicts today with such apparently illegal use of “AI” systems for target detection?
Taylor Kate Woodcock: Unfortunately, other cases of the use of AI systems for targeted attacks in various current conflicts, including in Iran, seem to confirm our findings. There we also see that the use of AI systems changes the target selection processes in such a way that this leads to ahttps://opiniojuris.org/2026/07/17/ai-enabled-targeting-and-the-structur...Deformation of the protective measures required by international lawleads. This can be seen, for example, in the reports of the United States bombing of the school in Minab in Iran, where the use of AI targeting systemshttps://www.reuters.com/world/middle-east/us-may-have-struck-iranian-gir...Problems with outdated information, with human verification and the protection of civilians can be made even worse.
Immediate shutdown
netzpolitik.org: What follows from your Lavender analysis?
Rainer Rehak: In our opinion, these systems should be shut down immediately. This requirement also applies to operations in Israel and Lebanon, and possibly also to other actors such as the United States. Secondly, we definitely need more transparency around such systems, especially if they are based on probabilities.
Thirdly, such systems are no longer allowed to be exported, as this is a common practice in Israel, as we seehttps://antonyloewenstein.com/books/the-palestine-laboratory-how-israel-...Work by the journalist Antony Loewenstein know. Fourth, we must look beyond the neat language of “precision strikes” and “targeted killings” and consider the material conditions and impacts on the ground.
If you don't do this, you're falling for a moral red herring and focusing on the wrong things. Israel has one of the most technologically advanced armies in the world and the use of Lavender and Co. with all its implications should be viewed as a conscious approach. Considering that Gaza is in what ahttps://www.tagesschau.de/ausland/asien/israel-gaza-kommissionsbericht-u...current United Nations Commission of Inquiryandhttps://en.wikipedia.org/wiki/List_of_humanitarian_and_human_rights_grou...relevant international organizationsrazed to the ground as genocide, Lavender is just a small piece of the puzzle, but perhaps a typical example of warfare in Gaza.
netzpolitik.org: Your article concludes by asking whether such systematic violations of international law should be considered war crimes that should be investigated by courts. Who could take such violations to court?
Taylor Kate Woodcock: Serious violations of international law can be tried as war crimes before international courts and tribunals, such as the International Criminal Court in The Hague, or by individual states that exercise so-called universal jurisdiction. This enables them to prosecute crimes that are so serious that they are considered crimes against all of humanity.
Currently, the International Criminal Court has issued an arrest warrant for Israeli Prime Minister Benjamin Netanyahu on charges of committing war crimes, namely the use of starvation as a weapon of war and the deliberate carrying out of attacks against the civilian population, as well as crimes against humanity, such as murder, persecution and other atrocities, committed since October 8, 2023 and until at least May 20, 2024. We believe that the use of AI in Gaza should be an integral part of this legal case that demonstrates systematic and serious violations of international law.
In addition to individual criminal responsibility, there are also frameworks in international law for the responsibilities of states, which are exercised by the International Court of Justice (ICJ), also in The Hague. In 2023, South Africa filed a lawsuit against Israel at the ICJ, accusing Israel of violating the Genocide Convention.
Other international organizations such as the UN have also commented on the genocide, which, while not legal action, still draws international attention to Israel's actions in the Gaza Strip. An example of this is thehttps://www.ohchr.org/en/press-releases/2025/09/israel-has-committed-gen...a United Nations commission of inquiry, which accuses Israel of genocide in the Gaza Strip.
In addition to these procedures, which take action against unlawful behavior after an event, international law also has an important preventive function. A large number of states are attempting to integrate artificial intelligence into their military capabilities and should take international law seriously. Although the use of Lavender by the Israeli army in Gaza took place in a very specific context, other states and regulatory authorities should still take a close look at the events to recognize that these systems require much more caution and transparency in order to comply with the obligations of international law.
Rainer Rehak: The Gaza conflict is unlikely to be solved through better and more advanced weapons systems, but only through political measures, by recognizing the realities and by addressing the core problems of the illegal occupation of Palestine and the fundamental rights of all people. If we don't focus our attention, we'll probably just see a change in the terrible technologies of war. International law emerged from the lessons of the Second World War, so we should take its principles seriously regardless of the actor.
netzpolitik.org: Thank you for the conversation and your time!
The work of netzpolitik.org is financed almost 100% from donations from our readers.
Become part of this unique community and support our public interest-oriented, advertising- and tracking-free journalismhttps://netzpolitik.org/spenden/?via=rss">nowwith a donation.
https://vg03.met.vgwort.de/na/5a2ba9626b16429c87c50b5b38759c2a"width="1" height="1" alt>
Read the full story at the source
Source: de.indymedia