This is a selective map of retrospective connections, not a linear genealogy and not a ranking of founders.
Keys 1–7 switch tabs · hover or focus a name for moreStart from contested problems rather than a canon of people. The same text may speak to more than one problem, and the same problem takes different forms in different settings.
A teaching map: links indicate shared problems, not direct influence. Hover or focus a name for the full reference and the register's reading note.
These focal texts provide an entry into the longer history. They are a deliberate teaching selection, not an authoritative canon or a claim that one tradition led to another.
Ringed marks are lecture anchors. Hover or focus one for its full reference and reading note. Dates identify publications, not the origin of a problem or tradition.
These are interpretive prompts for discussion, not summaries that exhaust the texts.
ELIZA made visible the gap between an apparently responsive conversation and the procedures that generate it—and the interpretive work people contribute themselves.
What changes when contemporary systems produce more convincing language?
Accounts of intelligent systems can mistake formal plans for the practical, contingent activity through which people make sense of a situation.
Whose situation disappears when an automated system treats a plan as complete?
Bias is not only a fault in a model. It can be inherited from social arrangements, introduced through technical choices, or emerge in a new context of use.
What would it take to locate responsibility rather than merely identify unequal outcomes?
As systems act in ways their designers cannot fully anticipate, familiar conditions for assigning responsibility can break down.
Does a “responsibility gap” name a missing person, a legal problem, or a problem in how institutions are organized?
Dates identify publications, not the origin of a problem.
The historical value of the earlier texts is not that they predict the present. They help distinguish recurring problems from new infrastructures, scales, and political arrangements.
A selective map of retrospective connections, not a linear genealogy.
How you organise the field decides which continuities you see. By lineage, a problem can be followed from 1950 to 2025. By track, the same 38 entries show what kind of activity made up the field — and everything before 2010 drops out of view.
The entries of tab v as a list: the handout table, and a way through the material without the chart.
| Date | Entry | Actors | Significance | Limits and contestation | Sources |
|---|---|---|---|---|---|
| 2011 | Operation LASER predictive policing Cases and controversies · Classification, bias & visibility Los Angeles, United States | LAPD; people subjected to policing; civil-rights advocates | LAPD's 2011 LASER programme made data-driven risk scoring an operational policing practice and later became a focal case for arguments that prediction can reproduce racially patterned enforcement data. | The programme's efficacy and causal role in disparate policing are contested; later reporting is stronger than contemporaneous public documentation. | Primary Secondary Secondary |
| 2012 | SyRI welfare-risk system begins operation Cases and controversies · Classification, bias & visibility · also Computer ethics & accountability Netherlands | Dutch government; municipalities; welfare recipients; privacy and anti-poverty groups | SyRI joined administrative datasets to identify welfare-fraud risks, bringing long-standing state classification into a data-intensive system later challenged by affected-rights and civil-society organisations. | Government descriptions framed SyRI as fraud prevention; critics disputed both legality and discriminatory effects. | Primary Secondary Secondary |
| 2013-06 | Snowden disclosures and the NSA archive Cases and controversies · Classification, bias & visibility · also Computer ethics & accountability Transnational | Edward Snowden; journalists; NSA and allied intelligence agencies; communications users | The disclosures made the scale of networked state surveillance publicly contestable, supplying a political context in which later data and AI accountability claims acquired urgency. | The archive documents programmes but not every later claim made in its name; national legal and political consequences varied. | Primary Secondary Secondary |
| 2014 | Amazon Mechanical Turk and the naming of crowdwork Institutions and publics · Automation & labor Transnational platform | Amazon; requesters; crowdworkers; labour researchers and organisers | Research and worker accounts made visible the dispersed human labour that platform-mediated data production and machine-learning development could present as automated. | MTurk is neither all data work nor representative of labour arrangements globally. | Primary Secondary Secondary |
| 2015 | Weapons of Math Destruction makes automated inequality a public vocabulary Scholarship and concepts · Classification, bias & visibility · also Computer ethics & accountability United States, widely read internationally | Cathy O'Neil; journalists; educators; affected publics | O'Neil's book translated concerns about high-stakes scoring into a widely usable account of feedback, opacity and inequality, helping make algorithmic harm a public issue beyond specialist venues. | Its broad category of 'WMD' can obscure differences among models, institutions and remedies. | Primary Secondary Secondary |
| 2016-05 | ProPublica publishes Machine Bias Cases and controversies · Classification, bias & visibility Florida, United States | ProPublica; Northpointe/Equivant; courts; defendants; statisticians | ProPublica's investigation turned competing measurements of racial disparity in recidivism scoring into a public controversy over what fairness can mean when base rates differ. | Northpointe disputed ProPublica's analysis; the episode should teach incompatible metrics, not a settled single diagnosis. | Primary Secondary Secondary |
| 2016-09 | FAT* conference formation Institutions and publics · Computer ethics & accountability · also Classification, bias & visibility International, initially United States | Conference founders; ACM; computer scientists; social scientists; civil-society participants | The Fairness, Accountability, and Transparency conference institutionalised a recurring venue where computer science, law and social research could contest the social consequences of algorithmic systems. | Conference participation, disciplinary authority and geographic access remained uneven. | Primary Secondary Secondary |
| 2017-02 | Algorithmic Justice League formed Institutions and publics · Classification, bias & visibility United States with transnational public reach | Joy Buolamwini; artists; advocates; communities affected by coded bias | The Algorithmic Justice League used research, art and advocacy to make coded bias publicly intelligible and to contest who counts as an expert on algorithmic harm. | Organisational self-description documents mission, not impact; media attention is concentrated in the United States. | Primary Secondary Secondary |
| 2017-11 | Toronto Declaration on machine learning and human rights Governance and infrastructure · Classification, bias & visibility · also Computer ethics & accountability Toronto, international | Access Now; Amnesty International; human-rights organisations; scholars | The Toronto Declaration articulated machine-learning discrimination as a human-rights obligation for states and private actors, widening the frame beyond technical bias correction. | A declaration is normative advocacy, not proof of implementation or legal force. | Primary Secondary Secondary |
| 2017-11 | AI Now Institute launches Institutions and publics · Computer ethics & accountability New York, United States | Meredith Whittaker; Kate Crawford; NYU; researchers; advocates | AI Now's launch created a durable research-and-policy institution that treated AI's social consequences as a cross-sector public question rather than an internal engineering matter. | It is a US-based institution and cannot stand for a global public sphere. | Primary Secondary Secondary |
| 2018-02 | Gender Shades audit Scholarship and concepts · Classification, bias & visibility United States, commercial systems marketed globally | Joy Buolamwini; Timnit Gebru; IBM; Microsoft; Face++; benchmark subjects | Gender Shades demonstrated large intersectional performance disparities in commercial gender classification and made audit design, benchmark composition and vendor accountability central public questions. | The study audits gender classification rather than all facial-recognition uses; vendors changed systems and benchmarks after publication. | Primary Secondary Secondary |
| 2018-04 | Google workers oppose Project Maven Cases and controversies · Computer ethics & accountability · also Automation & labor United States | Google employees; organisers; Google; US Department of Defense | Employee organising against Project Maven made the military uses of corporate AI a workplace and public-governance conflict, culminating in Google's decision not to renew the contract. | The scope of the workers' causal effect and Google's later defence contracts remain debated. | Primary Secondary Secondary |
| 2018-09 | Aadhaar judgment limits mandatory uses Cases and controversies · Classification, bias & visibility India | Supreme Court of India; UIDAI; petitioners; welfare recipients; civil-society organisations | India's Supreme Court upheld Aadhaar's constitutional basis while limiting several mandatory linkages, making biometric infrastructure a major site where welfare inclusion, exclusion and privacy were publicly litigated. | The judgment did not resolve ongoing exclusion, data-security or political disputes; Aadhaar is not simply an AI system. | Primary Secondary Secondary |
| 2018-12 | Datasheets for Datasets Governance and infrastructure · Computer ethics & accountability · also Classification, bias & visibility International research and industry | Timnit Gebru and co-authors; dataset creators and users | Datasheets proposed structured documentation of datasets' motivation, composition, collection, uses and maintenance, shifting accountability upstream from model outputs to data practices. | Adoption is partial; documentation can become procedural compliance without changing extractive collection. | Primary Secondary Secondary |
| 2019-05 | San Francisco bans city use of facial recognition Governance and infrastructure · Classification, bias & visibility · also Computer ethics & accountability San Francisco, United States | San Francisco Board of Supervisors; police; civil-liberties groups; residents | San Francisco's ordinance prohibited city departments' use of facial recognition and required approval procedures for surveillance technology, making local procurement a governance site. | The ban was jurisdiction-specific, and its effectiveness depended on definitions, enforcement and state/federal systems outside city control. | Primary Secondary Secondary |
| 2019-05 | OECD AI Principles adopted Governance and infrastructure · Computer ethics & accountability OECD member and partner countries | OECD member states; policy makers; businesses; civil society | The OECD Recommendation consolidated a state-facing vocabulary of inclusive growth, human-centred values, transparency, robustness and accountability that later policy instruments repeatedly invoked. | Principles do not establish enforcement; participation and capacity are unequal across states. | Primary Secondary Secondary |
| 2019-10 | Montreal Declaration for Responsible AI Institutions and publics · Computer ethics & accountability Quebec, Canada, with international signatories | Université de Montréal; civil society; researchers; public participants | The Montreal Declaration staged public consultation around responsible AI and articulated principles including well-being, autonomy, justice and environmental sustainability. | Its consultation and representation claims require scrutiny; its principles have no direct legal force. | Primary Secondary Secondary |
| 2020 | CARE Principles for Indigenous Data Governance Scholarship and concepts · No pre-2010 lineage in the register Global Indigenous data-governance network | Global Indigenous Data Alliance; Indigenous data stewards and researchers | CARE recast data governance around Collective benefit, Authority to control, Responsibility and Ethics, challenging governance models that treat data chiefly as an individual-consent or open-resource issue. | CARE is not a universal substitute for local Indigenous governance; implementation must be community-specific. | Primary Secondary Secondary |
| 2020-02 | Dutch court finds SyRI legislation incompatible with human-rights law Governance and infrastructure · Computer ethics & accountability · also Classification, bias & visibility Netherlands | District Court of The Hague; claimants; Dutch state; welfare recipients | The Hague court held SyRI legislation insufficiently transparent and disproportionate under Article 8 ECHR, demonstrating that welfare-risk infrastructure could be challenged before its outputs were treated as routine administration. | The ruling focused on the legal framework and privacy proportionality; it did not establish a general judicial test for all algorithmic discrimination. | Primary Secondary Secondary |
| 2020-08 | Ofqual grading crisis Cases and controversies · Classification, bias & visibility · also Computer ethics & accountability United Kingdom | Students; teachers; Ofqual; UK government; journalists; campaigners | The cancellation of algorithmically standardised A-level grades after public protest showed that technical moderation could be politically rejected when it redistributed life chances through institutional proxies. | The system was statistical standardisation, not machine learning; it is included to widen AI ethics' adjacent algorithmic governance history. | Primary Secondary Secondary |
| 2020-12 | Google's dismissal of Timnit Gebru becomes an AI ethics labour conflict Institutions and publics · Automation & labor · also Classification, bias & visibility United States, global research community | Timnit Gebru; Google; Google AI Ethics researchers; petitioning researchers and workers | The public dispute over Gebru's departure from Google exposed how corporate control of research employment could delimit which critiques of large-scale AI were institutionally sustainable. | Accounts of the employment decision conflict, and public reporting cannot adjudicate all internal facts. | Primary Secondary Secondary |
| 2021 | Abeba Birhane and colleagues publish 'The Impossibility of Automating Ambiguity' Scholarship and concepts · No pre-2010 lineage in the register · also Classification, bias & visibility International | Abeba Birhane; Vinay Uday Prabhu; Emmanuel Kahembwe; dataset subjects | The study showed how ImageNet's labels and images carried harmful, ambiguous and racialised classifications, moving dataset critique from technical quality toward historical and political conditions of categorisation. | The paper focuses on a particular dataset; removal or relabelling cannot alone resolve wider data extraction. | Primary Secondary Secondary |
| 2021-04 | European Commission proposes the AI Act Governance and infrastructure · Computer ethics & accountability · also Classification, bias & visibility European Union | European Commission; Parliament; Council; civil society; firms; member states | The Commission's proposal made risk classification, conformity assessment and prohibited practices central to a major regulatory project, shifting accountability toward pre-market governance. | Risk categories, enforcement resources, lobbying and the treatment of general-purpose AI were intensely contested. | Primary Secondary Secondary |
| 2021-11 | UNESCO Recommendation on the Ethics of Artificial Intelligence Governance and infrastructure · No pre-2010 lineage in the register · also Computer ethics & accountability UNESCO member states | UNESCO; 193 member states; experts; governments; civil society | UNESCO's Recommendation gave governments a broad ethical framework that explicitly included environmental impacts, gender, culture and data governance alongside human rights. | Unanimous adoption does not demonstrate domestic implementation, and state representation is not equivalent to affected-community representation. | Primary Secondary Secondary |
| 2022-02 | Meta content moderators in Kenya bring labour claims Cases and controversies · Automation & labor Kenya | Daniel Motaung; Sama workers; Meta; Sama; Kenyan courts and unions | Litigation and worker testimony brought the outsourced, traumatic labour sustaining platform safety into AI ethics debates about the supposedly automated governance of online content. | Proceedings and corporate responsibility claims developed over years; reports differ on employment relationships and remedies. | Primary Secondary Secondary |
| 2022-04 | DALL·E 2 and public generative-image disputes Cases and controversies · Automation & labor · also Machine intelligence & interaction Transnational | OpenAI; artists; photographers; users; journalists; dataset builders | The release and rapid public discussion of DALL·E 2 made questions of scraped cultural data, artistic labour and stereotyped representation newly central to popular accounts of generative AI. | Early commentary anticipated rather than proved specific legal or labour harms; systems and terms changed rapidly. | Primary Secondary Secondary |
| 2022-09 | LAION-5B scrutiny and removal practices Governance and infrastructure · Classification, bias & visibility · also Computer ethics & accountability Germany and transnational web | LAION; Common Crawl; data subjects; artists; journalists; researchers | Scrutiny of LAION-5B exposed how open research datasets could incorporate personal, sexual and copyrighted web material at a scale that made provenance and redress difficult. | Dataset content, removal processes and downstream model use changed; claims about any individual model require separate evidence. | Primary Secondary Secondary |
| 2022-11 | ChatGPT release accelerates public and institutional contestation Cases and controversies · Machine intelligence & interaction · also Automation & labor Transnational | OpenAI; teachers; students; writers; regulators; users | ChatGPT's public release rapidly moved foundation-model questions from specialist debate into schools, workplaces and mass media, intensifying disputes over authorship, reliability and access. | Usage figures and claims of educational transformation are often promotional or speculative; outcomes differ by setting and language. | Primary Secondary Secondary |
| 2023-01 | Artists bring copyright claims over image generators Cases and controversies · Automation & labor United States with transnational implications | Sarah Andersen; Kelly McKernan; Karla Ortiz; Stability AI; Midjourney; DeviantArt | The Andersen litigation made artists' objections to training-data extraction a formal legal conflict and helped establish generative-AI training as a contested issue of cultural labour and ownership. | Pleadings are allegations and rulings evolved; US litigation does not settle legal questions elsewhere. | Primary Secondary Secondary |
| 2023-03 | GPT-4 system card institutionalises frontier-model risk disclosure Governance and infrastructure · Computer ethics & accountability Transnational corporate infrastructure | OpenAI; external red-teamers; users; regulators | GPT-4's system card made red teaming and published model-risk documentation a visible part of foundation-model governance, while leaving firms substantial discretion over evidence and release decisions. | Corporate system cards are self-produced, selectively disclosed and not substitutes for independent audit. | Primary Secondary Secondary |
| 2023-05 | Writers Guild strike foregrounds generative AI and bargaining Institutions and publics · Automation & labor United States, global media impact | Writers Guild of America; writers; studios and streamers | The WGA strike placed generative AI within collective bargaining over credit, compensation and human creative work, showing that governance may arise through labour contracts as well as regulation. | The agreement applies to a bounded sector and does not resolve training-data questions or conditions outside union coverage. | Primary Secondary Secondary |
| 2023-10 | US Executive Order on AI Governance and infrastructure · Computer ethics & accountability United States | White House; federal agencies; NIST; companies; civil-society critics | The Executive Order directed federal agencies toward reporting, standards, civil-rights and worker-related actions, demonstrating executive coordination while relying heavily on existing agency authority and future implementation. | Executive orders can be revised or rescinded, and agency capacity and legal authority constrain effects. | Primary Secondary Secondary |
| 2023-12 | EU political agreement on the AI Act Governance and infrastructure · Computer ethics & accountability European Union | European Parliament; Council; Commission; civil society; industry | Political agreement on the AI Act showed that foundation models and biometric safeguards had become contested legislative objects rather than merely voluntary ethics topics. | A political agreement is not the operative Regulation; details changed before publication and implementation. | Primary Secondary Secondary |
| 2024-03 | +972 and Local Call report AI-assisted targeting in Gaza Cases and controversies · Computer ethics & accountability · also Classification, bias & visibility Gaza/Israel, contested reporting context | Israeli military; Palestinian civilians; +972 Magazine; Local Call; sources; human-rights investigators | Investigative reporting on AI-assisted target generation made military data systems a central and contested AI-ethics concern, raising questions about human judgement, verification and civilian harm under conditions of war. | Reporting relies partly on anonymous sources and is disputed by the IDF; claims must be attributed rather than presented as judicial findings. | Primary Secondary Secondary |
| 2024-05 | Council of Europe adopts Framework Convention on AI Governance and infrastructure · Computer ethics & accountability Council of Europe, open to non-members | Council of Europe; signatory states; civil society; companies in scope through state obligations | The Convention created the first international legally binding treaty framework explicitly linking AI activities to human rights, democracy and rule of law, while permitting substantial state discretion in implementation. | It does not itself create a uniform global regulator and has criticised public-security and national-defence exclusions. | Primary Secondary Secondary |
| 2024-08 | EU AI Act enters into force Governance and infrastructure · Computer ethics & accountability European Union | EU institutions; national authorities; providers; deployers; rights groups | Regulation (EU) 2024/1689 entered into force, turning years of ethics and policy debate into a staged legal regime whose effectiveness depends on implementation, standards and enforcement. | Most obligations apply on different dates; implementation, delegated acts and national capacity remain contested. | Primary Secondary Secondary |
| 2024-09 | California SB 1047 veto exposes contest over frontier-AI governance Cases and controversies · Computer ethics & accountability California, United States | California legislature; Governor Gavin Newsom; AI companies; safety advocates; open-source advocates | The veto of SB 1047 made disagreement over pre-deployment duties for powerful models publicly visible, including conflict over whether safety law would constrain or entrench major firms. | The bill was not enacted; participants disagreed sharply about its likely effects and underlying risk assumptions. | Primary Secondary Secondary |
| 2025-02 | AI Action Summit and the 'inclusive and sustainable AI' declaration Institutions and publics · No pre-2010 lineage in the register · also Computer ethics & accountability Paris, France, international | French government; participating states; international organisations; firms; civil society | The Paris summit made infrastructure, energy, public-interest AI and geopolitical coordination explicit themes of high-level AI governance, while exposing disagreement over the declaration and whose interests global cooperation serves. | Signatures and commitments do not prove redistribution of compute, data or decision-making capacity; some governments did not sign the declaration. | Primary Secondary Secondary |
Read from lecture/post2010/post2010_timeline_csv.csv when this file was built. No citation counts, rankings or quantitative encoding of importance.
What the visuals can show, what they cannot, and where every mark and note comes from.
This timeline is a teaching map, not a neutral inventory of the history of AI ethics. It selects events, institutions, texts and conflicts that are comparatively documentable in sources accessible to the research team. That selection makes some connections visible while leaving others thin, uncertain or absent. The most important use of this note is therefore methodological: it names the conditions under which a timeline can look more comprehensive, more coherent and more settled than the history it represents.
The entries were chosen because they reorganise a concept, make a harm politically consequential, establish an institution or governance practice, contest authority over technical futures, or connect cautiously to an older problem history. These criteria do not remove judgement. They privilege events that leave an accessible public record and that can be rendered in a short caption with a primary source and independent secondary support. Quiet accommodation, everyday repair, informal organising, refusal that never became public, and harms that were normalised rather than contested will be less visible.
Nor does a four-track structure describe four separate worlds. Scholarship can be a form of organising. Litigation can produce concepts. Journalism can create a public. Corporate documentation can be both a governance device and a public-relations intervention. The tracks are teaching aids for comparison, not a theory of causation. An entry's location on one track should not be read as a verdict about what mattered most.
The map is disproportionately shaped by English-language materials, open web archives, well-resourced legal systems and institutions that publish stable documents. A court decision, a major newspaper investigation, a conference archive or a formal policy text is easier to cite than oral testimony, community knowledge, materials in languages unavailable to the researchers, precarious worker communication or records held behind paywalls and security restrictions. This is an archival inequality, not evidence that the former matters more.
The geographic distribution of entries should be read in the same way. Including India, Kenya, the Netherlands, international Indigenous data-governance work, Palestine/Israel and UNESCO does not make the timeline global. It only prevents a narrower story from passing as universal. Large parts of Africa, Latin America, West Asia, South Asia, East and Southeast Asia, the Pacific, Eastern Europe and Indigenous territories remain underrepresented. So do linguistic communities whose experiences of training data, language technologies and platform moderation are poorly represented in English research infrastructure.
A further difficulty concerns security and danger. In border control, policing, war, migration, labour and welfare administration, people may be unable safely to document a system, challenge a decision or identify themselves publicly. Secrecy, non-disclosure agreements, inaccessible procurement systems and proprietary claims can make the infrastructure least open to scrutiny precisely where stakes are highest. Absence from the archive cannot be treated as absence from the history.
Some included systems are not AI in a narrow contemporary technical sense. SyRI and Ofqual, for example, belong because disputes about automated classification, statistical decision systems and administrative infrastructure shaped the conditions in which AI ethics became meaningful. Labelling every data system AI would erase important technical and institutional differences. Refusing to include adjacent systems would imply a false boundary between machine learning and the administrative systems from which it draws data, authority and harms.
The timeline therefore treats “AI ethics” as a contested historical formation rather than a fixed object. It asks how older problems were reassembled under changing technical and institutional conditions. It does not claim that an older debate directly caused a later one unless a source documents that connection. Similarity, recurrence and retrospective usefulness are not evidence of influence.
One sentence can describe a legal ruling, an audit or a worker action. It cannot settle the disagreements surrounding it. The COMPAS controversy includes disputes about methods and fairness criteria. Corporate departures involve conflicting accounts of employment decisions. Generative-AI lawsuits involve allegations, partial rulings and unresolved legal questions. Reporting on military AI can rely on anonymous sources and may be publicly contested. Governance principles may be ambitious yet non-binding. Laws may be enacted yet weakly enforced.
For this reason, captions distinguish what a source documents from an interpretation of its significance. Students should open the linked primary sources, compare independent accounts and ask what each source can support. A timeline should lead into evidence work, not replace it.
Once a sequence is displayed from 2010 to 2025, it can appear to culminate naturally in foundation models, international regulation or a handful of famous concepts. That appearance is misleading. Contemporary prominence is not proof that an event was decisive at the time, that it travelled across contexts, or that it deserves to represent a field. Citation counts, elite venues and corporate visibility are particularly poor proxies for political importance.
The timeline also risks converting critique into a canon of named texts and institutions. That would repeat the field's tendency to treat affected communities as sources of cases and researchers, firms or regulators as sources of theory. The map counters this only partially by including organising, journalism, worker disputes and litigation. It should be revised with community-produced sources, local histories, non-English materials and accounts of repair and refusal.
Treat every entry as a proposition open to revision. Ask whose account establishes the event, which actors are named as agents, what is missing from the available record, and whether the caption mistakes a formal commitment for material change. Add local cases and sources in the languages relevant to the class. Pair each high-profile event with a less visible infrastructure or labour history. Keep an exclusion log recording candidates that were rejected because sources were inaccessible, claims could not be verified, or the relation to AI was too weak.
The value of this timeline lies not in completeness. It lies in making its choices, evidentiary limits and political absences inspectable. A plural history of AI ethics must remain revisable, because the conditions under which its archives are made and its futures are contested remain unequal.
lecture/post2010/post2010_timeline_csv.csv when this file is built. Each has a primary source and two secondary sources. 32 are dated to the month; 6 to the year only, and those are placed mid-year.This file is generated. Edit the sources, then run python3 lecture/build_lecture_pack.py. Sources: the post-2010 CSV, lecture/pre2010_ai_ethics_history.html for the texts of tabs i–iv, lecture/post2010/What_this_timeline_cannot_show.txt, data/out/ancestors.json and reading/ancestor_claims.csv. It opens from disk and needs no network.
A lecture aid. Verify any quotation or expanded claim against the primary source before presenting it.