Submission 160
United Nations Special Rapporteur on human rights and counter-terrorism
Submitter group: Other.
An AI model identified the submitter group and wrote the summary, key points and recommendation summaries. Counts do not measure importance or merit.
Summary
Read the original submission for the full text.
Submits the Special Rapporteur's 2025 position paper on human rights while countering terrorism, written in an individual capacity and not necessarily reflecting an official UN position. It acknowledges AI's potential benefits, reviews uses in intelligence, surveillance, borders, courts, detention and the military, and recommends legal authorisation, prohibitions, due diligence, human control, oversight and remedies.
Key points
Quotes are the submitter’s own words. Text taken from documents, including scanned pages read by text-recognition software, can differ from the PDF. Check the original. Listed in the same order as on the submission’s page in the interactive guide.
States that certain State uses of AI in counter-terrorism can profoundly infringe human rights.
Critically, certain uses of AI systems by State authorities in countering terrorism can profoundly infringe human rights.
Argues that AI use in counter-terrorism, intelligence, law enforcement and military contexts must be authorised by precise legal provisions.
The development and use of AI systems in counter-terrorism must be authorised by law. Legal authority to use AI systems in law enforcement, intelligence and military contexts must conform to the principle of legality, such that permitted uses must be precise and specific, reasonably certain and foreseeable, and avoid vague and overbroad powers that are prone to abuse.
Argues that AI systems that cannot be used in compliance with human rights law must be prohibited.
AI systems must be prohibited where they cannot be used in compliance with human rights law.
Says oversight bodies need access to datasets, algorithms and audit trails to scrutinise AI use.
Oversight bodies must be given access to all relevant datasets, algorithms, stored data and audit trails, and have the capacity to review data retention and management practices and real-time digital operations, as well as inter-agency and cross-border data sharing.
Argues that indiscriminate AI-powered mass surveillance should be prohibited and targeted surveillance tightly constrained.
Indiscriminate AI-powered mass surveillance of the population must be prohibited. In relation to targeted individual or bulk surveillance, the law must precisely define the grounds for monitoring; and any interference must be necessary and proportionate in pursuit of a legitimate security aim, non-discriminatory, and subject to due process, independent oversight, judicial safeguards, and effective remedies.
Argues that in military settings AI must support rather than replace human decision-makers, to maintain chains of responsibility.
However, in order to maintain transparent chains of responsibility to respect IHL, AI systems must not replace human decision-makers and must be limited to supporting them.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Prohibit development, use, transfer and export of AI unable to meet human-rights standards or posing unacceptable rights risks; heighten regulation of permitted counter-terrorism AI.
It must be prohibited to develop, use and transfer and export AI systems that cannot meet human rights standards or pose an unacceptable risk to rights, and there must be heightened regulation of other AI systems that are permitted given the sensitivity of counter-terrorism contexts.
Require AI development and use in counter-terrorism to be authorised by law, with permitted uses precise, specific, reasonably certain and foreseeable, avoiding vague and overbroad powers.
The development and use of AI systems in counter-terrorism must be authorised by law. Legal authority to use AI systems in law enforcement, intelligence and military contexts must conform to the principle of legality, such that permitted uses must be precise and specific, reasonably certain and foreseeable, and avoid vague and overbroad powers that are prone to abuse.
Prohibit blanket military, national security, law enforcement or border management exemptions; permit tailored exceptions only when strictly necessary and proportionate to a legitimate aim, consistent with human-rights law and preserving effective remedies.
Blanket military, national security, law enforcement or border management exemptions must not be permitted; 193 any tailored exceptions must be strictly necessary and proportionate in pursuit of a legitimate aim and consistent with human rights law and must not deprive an individual of their right to an effective remedy.
Rely on effective State regulation rather than ordinary security, law enforcement or criminal justice rules, laissez-faire free-market approaches or industry self-regulation.
Given the risks to human rights, AI governance cannot be left to ordinary security, law enforcement or criminal justice rules, or to a laissez-faire free-market or industry self-regulation approach: effective State regulation is essential.
Deploy AI systems that interfere with human rights only where strictly necessary and proportionate in pursuit of a legitimate security aim, respecting all relevant rights.
States must only deploy AI-systems that interfere in human rights where they are strictly necessary and proportionate in pursuit of a legitimate security aim. Their use must respect all relevant rights, including equality and non-discrimination, privacy and personal data protection, freedoms of association, peaceful assembly, expression and religion, information and media freedoms, participation in public affairs, liberty, and the right to effective remedies and reparation.
Require public and private sector developers and deployers of AI systems to conduct rigorous human rights due diligence assessments throughout the lifecycle of such systems.
Public and private sector developers and deployers of AI systems must conduct rigorous human rights due diligence assessments throughout the lifecycle of such systems.
Prohibit human-rights-incompatible AI, including terrorism-risk assessment solely from profiling or personality traits without human assessment of objective, verified crime-linked facts, and facial databases built through untargeted, non-consensual image scraping.
AI systems must be prohibited where they cannot be used in compliance with human rights law. 194 Examples include where: 195 the risk of an individual committing a terrorist offence is assessed solely based on profiling or personality traits, including based on biometrics or sentiment analysis, absent human assessments based on objective, verified facts linked to criminal activity; facial recognition databases are sourced from untargeted, non-consensual scraping of images from the internet or surveillance footage; “real-time” remote biometric identification is used to indiscriminately surveil in public places; biometric systems infer sensitive attributes; emotions are inferred in certain sensitive settings; and decisions about military targeting are fully automated.
Prohibit human-rights-incompatible AI, including indiscriminate real-time public biometric identification, sensitive-attribute biometric inference, emotion inference in certain sensitive settings, and fully automated military targeting.
AI systems must be prohibited where they cannot be used in compliance with human rights law. 194 Examples include where: 195 the risk of an individual committing a terrorist offence is assessed solely based on profiling or personality traits, including based on biometrics or sentiment analysis, absent human assessments based on objective, verified facts linked to criminal activity; facial recognition databases are sourced from untargeted, non-consensual scraping of images from the internet or surveillance footage; “real-time” remote biometric identification is used to indiscriminately surveil in public places; biometric systems infer sensitive attributes; emotions are inferred in certain sensitive settings; and decisions about military targeting are fully automated.
Impose a moratorium on the use of an AI system whose human rights compliance is uncertain until it is demonstrated that it can be used compliantly.
Where there is uncertainty about the human rights compliance of a particular AI system, there must be a moratorium on its use until it is demonstrated that it can be used in a manner compliant with human rights.
Prohibit high-risk technology incapable of meaningful human supervision.
Effective human control must be exercised over AI systems, and high-risk technology that is incapable of meaningful human supervision must be prohibited.
Subject AI systems in counter-terrorism, including automated processing of personal data to evaluate individuals, to heightened regulation including enhanced transparency, data governance, technical documentation and record-keeping, risk and quality management, and human oversight.
AI systems in counter-terrorism, including those involving automated processing of personal data to evaluate individuals, should be subject to heightened regulation, which should include 196 enhanced transparency requirements; adequate data governance; provision of technical documentation and record-keeping to enable compliance; risk and quality management systems throughout the technology lifecycle; and assurance of human oversight.
Require prior judicial authorisation, with prompt post-facto authorisation permissible in emergencies, for particularly invasive AI systems such as remote biometric identification.
Particularly invasive AI systems, such as remote biometric identification, should be subject to prior judicial authorisation, with prompt post-facto authorisation permissible in emergencies.
Require developers to design AI systems to maximise transparency and explainability so operators can understand algorithmic processes, outputs and the sources, currency and weighting of underlying datasets.
Developers should be required to design AI systems to maximise transparency and explainability in how they function, thus enabling their operators to adequately understand and interrogate the functioning of algorithmic processes and their outputs, as well as the sources, currency and weighting of underlying datasets.
Prohibit the domestic or international sale or transfer of AI systems that cannot be used in compliance with human rights, and bar transfer of uncertain systems until verified as compliant.
AI systems which cannot be used in compliance with human rights must be prohibited from domestic or international sale or transfer. Where there is uncertainty about the human rights compliance of an AI system, it must not be transferred until it is established that it can be verifiably used in conformity with human rights.
Subject transfer of other AI systems to rigorous human rights diligence, including a prohibition where there is a credible risk they could be used to violate human rights in the place of deployment.
The transfer of other AI systems must be subject to rigorous human rights diligence, including a prohibition on transfer where there is a credible risk that they could be used to violate human rights in the place of deployment, including by targeting human rights defenders, journalists and other civil society actors.
Require businesses developing AI for counter-terrorism to conduct rigorous, comprehensive human rights due diligence impact assessments, including identifying biases and impacts on vulnerable, disadvantaged and marginalised groups.
States must require businesses or others developing AI systems for use in counter-terrorism to conduct rigorous, comprehensive human rights due diligence impact assessments, including to identify biases and impacts on vulnerable, disadvantaged and marginalised groups (including detainees, minorities, persons with disabilities, women, children, migrants, and LGBTIQ+ persons).
Engage all relevant stakeholders, including diverse civil society, in developer due diligence; periodically review and evaluate initial assessments.
Due diligence processes should engage all relevant stakeholders, including diverse civil society organisations.198 Initial assessments require periodic review and evaluation, given that an AI system’s functioning may necessitate adjustments in its performance.
Externally review and validate developer human rights assessments, and make them public.
States must require businesses or others developing AI systems for use in counter-terrorism to conduct rigorous, comprehensive human rights due diligence impact assessments, including to identify biases and impacts on vulnerable, disadvantaged and marginalised groups (including detainees, minorities, persons with disabilities, women, children, migrants, and LGBTIQ+ persons). Due diligence processes should engage all relevant stakeholders, including diverse civil society organisations.198 Initial assessments require periodic review and evaluation, given that an AI system’s functioning may necessitate adjustments in its performance. Assessments should be subject to external review and validation and be made publicly available.
Require investors in AI to undertake relevant due diligence.
Investors in AI, whether in research, capital infrastructure or commercial activities, must also undertake relevant due diligence.
Require businesses to devote sufficient personnel, expertise, capacity and resources to due diligence.
Businesses must devote sufficient personnel, expertise, capacity and resources to facilitate due diligence.
Impose effective and dissuasive financial and administrative penalties for failures to conduct adequate human rights due diligence.
There must be effective and dissuasive financial and administrative penalties for failures to conduct adequate human rights due diligence.
Ensure representative, ethical, accurate, current and relevant datasets; test and validate systems in simulated operational contexts, including automation bias; re-test modifications or repurposing, and establish quality assurance or management systems.
AI system developers and deployers should ensure that datasets are representative, ethically sourced, accurate, current, and relevant to the intended purpose. AI systems should be rigorously tested and validated in environments that simulate their operational context, including to assess risks such as automation bias (namely, a deployer’s over-reliance on AI-enabled outputs); and they should be re-tested whenever they are modified or their intended purpose changes. Quality assurance or management systems should be established.
Mitigate human-rights risks throughout AI lifecycles through embedded safeguards, terminating projects where risks cannot be adequately mitigated.
This includes by conducting thorough human rights due diligence assessments of AI systems throughout their lifecycle, including design, sale and transfer, implementation, training, deployment, and post-use; mitigating risks where feasible (including through embedding safeguards – “technology by design”) and terminating projects where risks cannot be adequately mitigated.
Ensure AI transparency and explainability through appropriate disclosure to users, regulators, investors, affected persons and the public.
Companies must further ensure the transparency and explainability of AI systems, including through appropriate disclosure to users, regulators, investors, affected persons and the public.
Have developers ensure diversity of personnel within AI design and engineering teams and provide diversity training for management.
Developers should ensure diversity of personnel within AI design and engineering teams and provide diversity training for management.
Hold regular inclusive stakeholder consultations, including with those most affected by AI systems, and put in place mechanisms to report potential biases.
Inclusive stakeholder consultations should be regularly held, including with those potentially most affected by AI systems, and mechanisms should be put in place to effectively report any potential biases identified.
Invest in research on AI human-rights implications, particularly rights interdependence and indivisibility, and the laws, policies, procedures and technical innovations needed for rights protection.
All stakeholders should invest in research into the human rights implications of the use of AI, including in counter-terrorism, in particular to develop a more nuanced understanding of how the interdependency and indivisibility of human rights are engaged. Research should also consider the laws, policies, procedures and technical innovations necessary to ensure that human rights are fully respected and protected in AI systems.
Advance collaborative responsible AI research prioritising explainability, bias mitigation, privacy-preserving technologies and data security, with ethical standards and protection of research subjects under privacy and data-protection law.
Collaboration between government agencies, academic institutions, and private sector entities should advance responsible AI innovation, including by prioritising explainable AI, bias mitigation, privacy- preserving technologies, and data security, while adhering to ethical research standards and protecting subjects in accordance with data protection and privacy laws.
Conduct forward-looking collaboration on emerging technologies, including regular horizon scanning and scenario planning, and encourage technology-neutral language in legal standards where feasible.
broad collaboration amongst stakeholders on new and emerging technologies should be forward-looking, including by conducting regular horizon scanning and scenario planning exercises, and encouraging, where feasible, technology-neutral language to legal standards and frameworks that may be applied consistently over time.
Require deployers to conduct human rights impact assessments before, during and after use of AI systems, to set boundaries, mitigate adverse effects or refrain from or terminate use.
Those who deploy AI systems must conduct human rights impact assessments before, during and after use, to establish and maintain clear boundaries for AI-assisted decision-making processes, mitigate potential adverse effects or refrain from using or terminate use of the system, and ensure alignment with international human rights law and national law.
Consider the whole legal and technological environment, including integration with other technologies, and engage diverse stakeholders in deployer due diligence.
Due diligence must holistically consider the legal and technological environment in which they would be embedded, 201 including their integration with other new technologies. Due diligence processes should engage all relevant stakeholders, including diverse civil society organisations.
Resource business due diligence and human-rights compliance proportionately to the nature and scale of AI use, including content moderation.
Business must devote sufficient personnel, expertise, capacity and resources to facilitate due diligence and human rights compliance, proportionate to the nature and scale of the use of AI, including in areas such as online content moderation.
Require AI to support human decisions, with effective qualified oversight, heightened scrutiny of elevated rights risks, and prevention and minimisation of unintended or unanticipated human-rights harms.
AI should support, not replace, human decision-making in counter-terrorism contexts. Qualified personnel must retain effective oversight of operational uses of AI, with heightened scrutiny where impact assessments have identified elevated risks to human rights. Human oversight should aim to prevent and minimise risks to human rights from the operation of AI systems, including from unintended or unanticipated consequences.
Require comprehensible explanations of AI operation, inputs, outputs and recommendations, replacing opaque algorithms with interpretable models.
AI systems must function in a manner that allows for comprehensible explanations of their operation, inputs and outputs and recommendations. Algorithms functioning in an opaque manner should be replaced with interpretable models that enable understanding of reasoning processes, facilitating transparency and accountability.
Invest in public programs and community workshops that demystify algorithmic decision-making and foster informed civic participation in AI governance discussions.
Governments should invest in public programs and community workshops that demystify algorithmic decision-making processes and fostering informed civic participation in AI governance discussions.
Provide comprehensive AI and human-rights training and regular professional development covering evolving technology, law, cultural sensitivity and decision-making, with ongoing evaluation and review.
Personnel operating AI systems require comprehensive training in both their technological capabilities and implications for human rights protection. Regular professional development programs, subject to ongoing evaluation and review, should address evolving AI technologies, legal requirements, cultural sensitivity, and decision-making responsibilities in complex operational environments.
Make training available to those involved in oversight and accountability processes, including legal professionals and judges.
Training should also be available for those involved in oversight and accountability processes and mechanisms, including legal professionals and judges.
Prohibit indiscriminate AI-powered mass surveillance, and require the law to precisely define the grounds for targeted or bulk surveillance, with necessity, proportionality, non-discrimination, due process, independent oversight, judicial safeguards and remedies.
Indiscriminate AI-powered mass surveillance of the population must be prohibited. In relation to targeted individual or bulk surveillance, the law must precisely define the grounds for monitoring; and any interference must be necessary and proportionate in pursuit of a legitimate security aim, non-discriminatory, and subject to due process, independent oversight, judicial safeguards, and effective remedies.
Require AI-enabled bulk electronic interception to have prospective independent authorisation identifying the categories of selectors, with each selector justified as necessary and proportionate and detailed records kept.
AI-enabled bulk electronic interception must be subject to prospective independent authorisation; the authorisation must identify the categories of “selectors” to be used; each selector must be justified as necessary and proportionate; and detailed records must be kept.
Provide independent supervision and review of bulk surveillance with end-to-end safeguards assessing necessity and proportionality at each stage.
There must be independent supervision and review of bulk surveillance with “end-to-end safeguards” to assess necessity and proportionality at each stage.
Regulate online content moderation in accordance with the rights to freedom of expression, association, peaceful assembly, participation in public affairs, and non-discrimination.
States must regulate online content moderation in accordance with the rights to freedom of expression, association, peaceful assembly, participation in public affairs, and non-discrimination.
Restrict expression only under legality, necessity, proportionality and legitimate-aim requirements, consistent with the ICCPR, Rabat Plan and the Special Rapporteur’s incitement recommendations.
Expression may only be restricted, whether by States or private actors, in accordance with the requirements of legality, necessity, proportionality and legitimate aim, consistent with articles 19 and 20(1) of the ICCPR,202 the Rabat Plan of Action against hate speech, and the recommendations on incitement to terrorism of the Special Rapporteur on the Promotion and Protection of Human Rights and Fundamental Freedoms while Countering Terrorism.203
Require intent to incite terrorism as properly defined under international best practice, and reasonable probability or real risk that a terrorist act will occur.
The latter requires that there must be an intention to incite terrorism, as properly defined according to best practice international standards, as well as a reasonable probability or real risk that a terrorist act will be committed.
Focus interventions on pluralistic participation and viewpoints and accessible public-interest information; use AI to improve public information access, amplify marginalised voices and enable civic participation.
Interventions must focus not merely on suppressing harmful content but on promoting pluralism of participation and view-points and ensuring public interest information remains available and accessible. As the Global Digital Compact reaffirms, States must “refrain from imposing restrictions on the free flow of information and ideas that are inconsistent with obligations under international law”, 89 including the right to freedom of expression and opinion. AI should be harnessed to bolster democracy by improving access to public interest information, amplifying marginalized voices, and enabling civic participation.90
Ensure technology companies publicly disclose human rights due diligence and impact assessments, with heightened obligations in conflict and high-risk situations.
Further, States must ensure that technology companies undertake and publicly disclose human rights due diligence and impact assessments, with heightened obligations in conflict and high- risk situations.
Provide adequate, trained, unbiased human moderation and AI oversight; ensure algorithmic transparency; monitor mitigation effectiveness; publish moderation policies and reports; maintain trained human-rights staff; and provide effective internal appeals.
Companies should also provide adequate, well-trained, unbiased human content moderation and oversight of AI content moderation; ensure algorithmic transparency; monitor the effectiveness of mitigating actions; publish moderation policies and detailed moderation reports; maintain adequate and well-trained in-house human rights staff; and provide effective internal appeal mechanisms.204
Maintain multilingual and culturally diverse expertise for online content moderation.
Companies should also provide adequate, well-trained, unbiased human content moderation and human oversight of AI content moderation; ensure algorithmic transparency; monitor the effectiveness of mitigating actions; publish their moderation policies and detailed moderation reports; maintain adequate and well-trained in-house human rights staff and multilingual and culturally diverse expertise; and provide effective internal appeal mechanisms.
Have investors conduct investment due diligence and use their leverage to protect human rights.
Investors should conduct due diligence on investments and use their leverage to protect human rights.
Require meaningful human oversight and border-AI safeguards for privacy, data protection, non-refoulement, asylum, nationals’ movement, non-nationals’ entry and expulsion, family unity, non-discrimination, liberty, expression, association and peaceful assembly.
Any use of AI in border management must provide meaningful human oversight and safeguards for the rights to privacy and data protection, non-refoulement and asylum, freedom of movement of nationals (to enter and leave one’s own country), freedom of non-nationals from arbitrary expulsion or arbitrary denial of entry, the right to family unity, non-discrimination, liberty, and freedoms of expression, association and peaceful assembly.
Require border authorities to ensure transparency, explainability, dataset accuracy and non-discrimination in algorithms, robust data security, avoidance of over-collection, and rigorous safeguards on cross-border data sharing.
Border authorities must ensure transparency, explainability, dataset accuracy and non-discrimination in algorithms and implement robust data security. Unnecessary and disproportionate over-collection of data must be avoided, and there must be rigorous safeguards on cross-border data sharing.
Use behavioural biometrics and linguistic inconsistency analysis only after reliability is validated; safeguard against facial-recognition inaccuracies and bias across demographic groups.
New technologies such as behavioural biometrics and linguistic analysis of narrative inconsistencies should not be used unless their reliability is validated. Authorities must carefully safeguard against inaccuracies and bias in facial recognition systems across demographic groups.
Ensure judges govern judicial AI, with context-specific use respectful of local needs, languages, culture and legal traditions.
The use of AI in judicial systems must be governed by judges, context-specific, and respectful of local needs, languages, culture and legal traditions.
Ensure chain of custody, empirical validation and expert explanation for AI evidence; rigorously test it, prevent discrimination, exclude illegally obtained evidence under ordinary admissibility rules, and ensure effective challenges and judicial control.
Consequently, care is needed to ensure chain-of- custody of AI-enabled evidence processes, empirical validation, and expert explanation. AI evidence must be rigorously tested, non-discriminatory, subject to ordinary admissibility requirements (including to exclude evidence illegally obtained through AI systems), capable of being effectively challenged, and subject to judicial control. 116
Empirically validate risk-assessment and behavioural-insights tools, avoid automatic or undue evidentiary weight, and holistically consider all relevant factors, including mitigation and mental health.
Yet, all risk assessment and “behavioural insights” tools must be empirically validated and not given automatic or undue weight as probative of a person’s risk or the need to adopt a particular counter-terrorism measure in response, and all relevant factors, including mitigating circumstances and mental health, must be holistically considered. 120
Have judiciaries adopt guidelines on the use of AI to ensure respect for the right to fair trial, judicial independence, non-discrimination, liberty and other affected rights.
Judiciaries should adopt guidelines on the use of AI (see also UNESCO Guidelines for the Use of AI Systems in Courts and Tribunals 2025) to ensure respect for the right to fair trial (including equality of arms, legal representation and the presumption of innocence), judicial independence, non-discrimination, liberty and other affected rights.
Publish key information about judicial AI systems to enable legal challenges and civil society oversight.
Key information about judicial AI systems should be published, to enable legal challenges and civil society oversight.
Subject AI evidence, including digital and forensic evidence, to the ordinary rules of admissibility, disclosure and challenge.
AI evidence, including digital and forensic evidence, must be subject to the ordinary rules of admissibility, disclosure and challenge.
Never use predictive analytics as a basis for prosecution or detention, and stringently control its use in civil or administrative proceedings to restrict rights.
Predictive analytics should never be used as a basis for prosecution or detention and should be stringently controlled when used in civil or administrative proceedings to restrict rights.
Ensure detention AI does not adversely affect conditions, treatment or discipline, and retain sufficient direct human contact with detainees.
States must ensure that any AI systems in detention do not adversely affect the conditions of detention and the treatment and discipline of detainees, and that detention authorities retain sufficient direct contact with detainees.
Keep detention-eligibility AI subordinate to humans, with individual necessity and proportionality, judicial review, transparency, explainability and non-discrimination in algorithms and datasets.
The use of AI decision-support systems in determining eligibility for detention must remain subordinate to human decision-making, be consistent with the requirements of the right to liberty, including the individual necessity and proportionality of detention and judicial review, and ensure transparency, explainability and non- discrimination in relation to algorithms and datasets.
Maintain legal frameworks governing data collection, access, processing, sharing, transfer and storage for counter-terrorism so that practices are safe, secure, necessary and proportionate.
Legal frameworks must govern and ensure that data collection, access, processing, sharing, transfer, and storage practices are safe, secure and necessary and proportionate for legitimate counter-terrorism purposes, in accordance with international law.
Require public authorities and intelligence agencies to implement data minimisation, purpose limitation and secure erasure procedures, complying with data protection and privacy legislation.
Public authorities and intelligence agencies should implement data protection principles such as data minimization, purpose limitation requirements, and secure erasure procedures while ensuring compliance with data protection, privacy legislation and adherence to human rights standards that safeguard other interdependent rights.
Notify individuals when their personal data will become part of a dataset used by an AI system.
Individuals should be notified when their personal data will become part of a dataset used by an AI system.
Empower individuals and groups, as a general rule, to give and withdraw informed consent to the use of their data and to choose how it is used.
As a general rule individuals and groups should also be empowered to consider, give and withdraw their informed consent to the use of their data and to choose how their data is used.
Implement multi-layered security architectures protecting counter-terrorism AI systems against external threats and internal misuse, including adversarial attacks, data poisoning and system manipulation.
AI systems in counter-terrorism contexts present attractive targets for adversarial attacks, data poisoning activities such as the corruption of training data, and system manipulation. Multi-layered security architectures must protect against both external threats and internal misuse while maintaining system integrity and operational continuity.
Integrate equality and non-discrimination safeguards throughout AI development and use, including continuing due diligence; refrain from discriminatory state uses and regulate private actors against discrimination.
States must therefore integrate safeguards to ensure its use is consistent with the rights to equality and non-discrimination. International treaty and customary law prohibit discrimination on prohibited grounds, such as on the basis of race, colour, sex, ethnicity, age, language, religion, political or other opinion, national or social origin, disability, property, birth or other status. Non-discrimination and equality should be addressed within each phase of development of an AI system, including in its design, development, training and use, 164 including through ex ante and continuing due diligence assessments by States and business. 165 States must refrain from discriminatory uses of AI but also regulate private actors to protect against discrimination.
Disclose model specifications, algorithms, data sources and known, foreseeable or unknown limitations to affected persons, oversight bodies, regulators and the public; enforce disclosure exceptions for proprietary or classified algorithms.
Further, there must be sufficient disclosure to affected persons, as well to relevant oversight and regulatory bodies and the public, of all relevant elements of the AI-system, including model specifications and algorithmic configurations, data sources, known or foreseeable limitations, as well as what may be unknown. Where algorithms are proprietary or classified, enforceable exceptions enabling adequate disclosure must be made.
Review oversight frameworks and establish new transparency and accountability mechanisms where emerging technologies may pose high human-rights risks.
In situations where emerging technologies may pose a high risk to human rights, existing oversight frameworks must be reviewed and new mechanisms to enhance transparency and accountability must be established.
Require regular algorithmic audits for bias, accuracy and rights impacts, with independent oversight bodies evaluating training data and performance across groups.
Regular algorithmic audits must examine AI systems for bias, accuracy, adverse impacts and interferences with human rights. Independent oversight bodies should evaluate training data quality, machine learning model performance across different groups, and potential discriminatory outcomes, particularly regarding racial, religious, and ethnic profiling in surveillance and threat assessment applications.
Build record-keeping and audit trails into AI systems and maintain public registers of AI systems used in counter-terrorism.
The design of AI systems should incorporate adequate record-keeping capabilities, including automatic recording, in order to identify relevant events and risks and enable an audit trail. Public registers should also be maintained of AI systems used in counter-terrorism.
Have independent review bodies, legislative committees and judicial oversight mechanisms monitor AI use, investigate complaints and ensure compliance with the law and human rights while respecting confidentiality safeguards.
Independent review bodies, legislative committees, and judicial oversight mechanisms should monitor the use of AI systems, investigate complaints, and ensure compliance with the law and human rights while respecting confidentiality safeguards.
Give oversight mechanisms guaranteed independence, sufficient human, financial and technical resources including AI expertise, and a broad mandate with adequate powers.
Oversight mechanisms must have guaranteed independence, sufficient human, financial and technical resources (including relevant AI expertise), and a broad mandate with adequate powers.
Consider establishing a dedicated AI oversight body or regulator covering counter-terrorism, law enforcement, intelligence and security, border management and the military.
States should consider establishing a dedicated AI oversight body or regulator with competencies covering counter-terrorism, law enforcement, intelligence and security, border management, and the military.
Give oversight bodies access to all relevant datasets, algorithms, stored data and audit trails, and capacity to review data retention and management, real-time operations and data sharing.
Oversight bodies must be given access to all relevant datasets, algorithms, stored data and audit trails, and have the capacity to review data retention and management practices and real-time digital operations, as well as inter-agency and cross-border data sharing.
Require State and private actors who deploy AI to disclose transparently when and how AI is used and publicise information about complaints and remedial processes.
State and private actors who deploy AI should transparently disclose when and how AI is used and publicise information about complaints and remedial processes.
Require companies to establish, maintain and adequately resource accessible and effective complaints mechanisms for rights violations resulting from their AI systems.
Companies should establish, maintain and adequately resource accessible and effective complaints mechanisms for remedying violations of human rights resulting from their AI systems.
Ensure any individual who believes their rights were infringed by an AI system can bring a complaint to an independent oversight institution, such as an ombudsman, national human rights institution or a court.
Any individual who believes that their rights have been infringed by the use of an AI system must be able to bring a complaint to an independent oversight institution, such as an ombudsman, national human rights institution or a court.
Ensure victims of unlawful actions have recourse to an institution that can provide a binding, effective remedy, including full reparation.
Victims of unlawful actions, including rights violations, must have recourse to an institution that can provide a binding, effective remedy, including full reparation for the harm suffered.
Ensure due process in remedy procedures, including disclosure about algorithms, data sources and potential biases, access to legal representation and legal aid.
Procedures must ensure due process, including sufficient disclosure concerning algorithms, data sources and potential biases and access to legal representation. Justice must be accessible, including through legal aid in accordance with international law.
Have international cooperation mechanisms address jurisdictional challenges in AI-enabled counter-terrorism, with information-sharing protocols that safeguard individuals' rights across legal systems.
International cooperation and coordination mechanisms should address jurisdictional challenges in AI-enabled counter-terrorism operations. Standardized protocols for information sharing, joint operations, mutual legal assistance and extradition should incorporate safeguards for individuals’ rights across different legal systems while maintaining operational effectiveness.
Do not arbitrarily exclude the privacy and other rights of non-citizens abroad from legal protection.
The privacy and other rights of non-citizens abroad should not be arbitrarily excluded from legal protection.
Establish specialised AI-security governance bodies within existing institutions to harmonise technical standards, operational procedures and human-rights safeguards across jurisdictions.
Institutional frameworks should be further strengthened by establishing specialised bodies within existing structures dedicated to AI governance in security contexts. These institutions should work to harmonize technical standards, operational procedures, and human rights safeguards across jurisdictions, facilitating mutual support and cooperation in countering terrorism.
Consider multilateral treaties on cross-border AI surveillance and decision-making data transfers; any new arrangements should strengthen transparency, explainability, accountability and other human-rights safeguards.
Multilateral treaties addressing cross-border data transfers from AI-enabled surveillance and automated decision-making in counter-terrorism contexts could nonetheless be considered. Such instruments could assist in reconciling complex and divergent data protection and privacy regimes, constitutional provisions, and cultural perspectives on the acceptable limits of surveillance while maintaining operational effectiveness against increasingly sophisticated threat actors operating across jurisdictions. Any new arrangements should not merely maintain existing levels of human rights protection but strengthen safeguards, including transparency, explainability and accountability.
Develop certification frameworks to enable secure information exchange between certified systems while ensuring compliance with human rights, including privacy and data protection.
Certification frameworks should be developed to enable secure information exchange between certified systems while ensuring compliance with human rights, including privacy and data protection.
Develop standards to verify and authenticate classified AI data, ensuring disclosure where necessary to protect rights, including fair hearings and challenges to evidence.
Standards and procedures must therefore be developed to verify and authenticate classified data used in AI systems, while ensuring that adequate disclosure is afforded where necessary to protect human rights, such as in the contexts of the fair hearing right to know and challenge the evidence, whether in criminal trials or when seeking remedies for violations such as of the right to privacy.
Prioritise law-enforcement AI certification standards covering data quality, purpose specification, transparency, bias mitigation and security, with mutual recognition for secure information exchange.
International technical standards organizations can assist by prioritising the development of certification frameworks for AI systems employed in law enforcement operations. These standards should address data protection requirements such as data quality assurance and purpose specification, in addition to other requirements such as algorithmic transparency, bias mitigation and data security, while establishing mutual recognition of standards that enable secure information exchange between certified systems.
Ensure AI systems follow universal accessibility principles; prioritise AI literacy, institutional capacity and technology transfer in competency programs for equitable counter-terrorism participation under rigorous human-rights standards.
Critically, developers must ensure that systems are designed in accordance with universal accessibility principles, accommodating users irrespective of age, gender, ability, or other distinguishing characteristics. 191 Programs to develop competencies should prioritise AI literacy, strengthening institutional capacity, and technology transfer mechanisms to ensure equitable participation in global counter-terrorism efforts while maintaining rigorous human rights standards across all jurisdictions. 192
Strengthen human-rights-based AI regulation through international stakeholder collaboration and dialogue, clarifying transparency, accountability, due diligence and human-rights norms throughout AI development and deployment.
The international community including States, international organisations and civil society should work together to further strengthen human rights-based regulation of the future development and use of AI. Greater collaboration through engaging all relevant stakeholders can build on the existing normative frameworks to clarify and enhance the oversight of AI, especially to provide the necessary guidance in the areas of transparency, accountability, human rights due diligence and the more precise and nuanced diffusion of human rights norms throughout all phases of the AI development and deployment lifecycle.
Do not share data where it is likely to be used for the purpose of violating human rights.
Data should not be shared where it is likely that it will be used for the purpose of violation of human rights.
For UN entities, address AI capacity-building and technical-assistance human-rights implications, including biometrics and borders, particularly for states with demonstrated counter-terrorism rights violations.
All relevant United Nations entities must address the human rights implications of providing capacity-building and technical assistance capacity in relation to AI systems (including in relation to biometrics and border management tools), particularly regarding States with demonstrated records of human rights violations in counter-terrorism. 212
For UN capacity-building and technical assistance, establish moratorium and suspension protocols extending beyond situations covered by the UN Human Rights Due Diligence Policy.
Both moratorium and suspension protocols on capacity-building and technical assistance should be established, not limited to situations covered by the Human Rights Due Diligence Policy on United Nations Support to Non-United Nations Security Forces.
Have Member States and international organisations ensure processes to strengthen global AI governance preserve compliance with international human rights, humanitarian and refugee law, including through specific initiatives for counter-terrorism and security uses.
Member States and international organisations should ensure that processes to strengthen global governance of AI, including through the Global Dialogue on AI Governance and the Independent International Scientific Panel on AI, preserve and strengthen compliance with international human rights law, international humanitarian law and international refugee law, including through specific initiatives to particularise the effective application of these norms where AI systems are used in counter-terrorism or other security contexts.
Consider how international humanitarian and human-rights law apply to military AI and implement existing laws for compliance, oversight, governance, accountability and remedies.
States must consider how these legal frameworks specifically apply to their AI capabilities and how existing laws can be effectively implemented to ensure compliance, oversight, governance and accountability and remedies.
Provide civilian oversight of military AI systems.
While acknowledging the expertise of militaries, civilian oversight of military AI systems is also imperative. 132
Keep AI decision-support systems understandable and preserve sufficient time and space for human deliberation, assessment and verification in military decisions.
In order to maintain transparent chains of responsibility and thus compliance with IHL and human rights law, AI DSS must remain understandable to humans and preserve sufficient time and space for human deliberation, assessment and verification. 137
Urgently negotiate an autonomous-weapons treaty by 2026, limiting locations, timing, duration, target types and force scale, with effective human supervision and timely intervention and deactivation.
States should urgently negotiate by 2026 a treaty regulating autonomous weapons systems, as advocated by the Secretary-General and the ICRC, 213 including to limit where, when and for how long they are used, the types of targets they strike and the scale of force used, and to ensure effective human supervision and timely intervention and deactivation.
Endorse ICRC military AI decision-support recommendations, preserving human control and judgment over life, liberty and dignity, and ensuring lawful, responsible use under international humanitarian law.
Endorse the ICRC’s recommendations on AI-enabled decision support systems 214 In particular, States and parties to armed conflicts must ensure that human control and judgement are preserved in decisions that pose risks to the life, liberty, and human dignity (ICRC Recommendation 1) and ensure lawful and responsible use in compliance with IHL (R11);
For military AI decision support, maximise transparency and explainability, mitigate discriminatory bias, review weapons-system uses legally, assess all reasonably available information, and protect detainee treatment and conditions.
systems should be designed to maximise transparency and explainability (R2); developers and users should mitigate bias to prevent discriminatory outcomes (R4); legal reviews should be conducted wherever AI DSS forms part of a weapons system (R7); users should assess information from all sources reasonably available when drawing on AI DSS outputs (R13); and the use of AI DSS in detention operations must not adversely affect the treatment of detainees and the conditions of detention (R14).