Users on X are asking Grok to imitate Israel’s reported Lavender targeting system by assigning numerical “concern” scores to accounts based on their public posts, including profiles that regularly criticize Israel or express support for Palestinians.
In one example, a user instructed Grok to examine an entire X account through the “Israel’s Lavender” framework and provide only a score between 1 and 100.
Grok instead produced an assessment of the account’s activity, noting posts critical of Israel, including accusations of genocide and opposition to US aid for Israel.
It characterized the account as belonging to a US-based content creator involved in political commentary and said it found no apparent militant connections, relevant location information or operational indicators.
The response ultimately described the account as presenting negligible direct concern.
Grok Is Not Running Israel’s Lavender System
The example does not mean that Grok has access to Israel’s actual Lavender system or is operating Israeli military software.
X says Grok can search public posts on the platform and use information from X when generating responses. This allows users to ask the chatbot to examine publicly available activity on an account.
Any “Lavender” score generated in response to such a prompt is therefore a Grok-generated simulation, based on the user’s instructions and publicly available descriptions of the reported system.
Such an assessment should not be interpreted as an intelligence finding. X also warns users that Grok can produce incorrect information with confidence and may fail to account for important context.
What Is the Lavender System?
Lavender was widely reported in 2024 following investigations by +972 Magazine and Local Call, which cited interviews with six Israeli intelligence officers.
The reports described Lavender as an AI-assisted system used to process large volumes of information and identify people suspected of belonging to the military wings of Hamas or Palestinian Islamic Jihad.
The investigation reported that the system flagged tens of thousands of people during the Gaza war.
The Israeli military disputed significant elements of those reports. The IDF said its information systems support human analysts and denied that an AI system independently decides whether a person is a militant.
The military also said analysts are required to independently verify potential targets and that human personnel review strike decisions to ensure they comply with international law.
Subsequent reporting has described Lavender as generating numerical assessments of suspicion using multiple intelligence sources. However, the system’s precise methodology remains classified and has been the subject of public dispute.
AI Risk Scores Based on Social Media
The Grok example highlights how a general-purpose AI model with access to public social-media content can be instructed to imitate the concept of a military risk-assessment system.
That does not establish that X or xAI is using Lavender to monitor or classify pro-Palestinian users.
Grok does not have the classified intelligence, surveillance information, methodology or verified decision-making rules reportedly associated with Lavender. Its responses are generated from the information available to the model and the instructions contained in a user’s prompt.
As a result, a numerical score produced by Grok should be understood as an AI-generated response to a hypothetical framework, rather than an authentic Lavender assessment or evidence that the account has been evaluated by Israeli authorities.
The distinction is particularly important when AI-generated scores are applied to real people, as public posts can lack context and do not by themselves establish a person’s affiliations, intentions or activities.
