Anu Bradford

Who she is

Affiliation: Henry L. Moses Professor of Law and International Organization, Columbia Law School (chair since 2014); Director, European Legal Studies Center; Senior Scholar, Chazen Institute (Columbia Business School); nonresident scholar, Carnegie Endowment; Faculty Associate, Berkman Klein Center (Harvard). Role check run 2026-09-21 — new since v1: Special Advisor to European Commission Executive Vice-President Henna Virkkunen (Tech Sovereignty, Security and Democracy), named in 2025 (Berkman profile, updated 2026-08-04; the Columbia bio lists "Special Advisor for the Executive Vice President of the European Commission"). Appointment month and terms (paid or not) not found in a Commission document this cycle. Elected to the Finnish Academy of Science and Letters, 2026 (Columbia bio). Finnish-born; author of The Brussels Effect (OUP 2020) and Digital Empires (OUP 2023-09; FT best books of 2023; 2024 Stein Rokkan Prize). Identity note: a search summary credited her with a Jean Monnet Chair (2026–29) at City St George's, London; that chair is Elaine Fahey's — Bradford only gave a book talk under it.
One-line position: The EU's rules travel the world because its market is large and companies would rather run one standard than two — the "Brussels Effect" — so EU digital law becomes the global default whether anyone votes for it or not. Revised by its author, 2026-03: "the high-water mark of Europe's regulatory leadership is behind us" — Europe's resolve "has wavered under US pressure and internal doubts," and in her telling the retreat, not the rulebook, is the error (Capgemini).

Discipline & technical bet

Comparative and international law crossed with political economy; a non-technical thinker with no technical wager. Her bet is institutional: a rights-driven regime beats the US market-driven and Chinese state-driven regimes on legitimacy and staying power, and Europe's tech lag comes from things regulation cannot fix — no true single digital market, thin capital markets, punitive bankruptcy law, talent flows ("The False Choice Between Digital Regulation and Innovation," Northwestern University Law Review 119:377, 2024; restated 2026-03). The revealing 2025–26 move: she went inside, as Special Advisor to the Commissioner whose portfolio produced the Digital Omnibus (proposed 2025-11-19) — a wager that the rollback can be bounded from within rather than denounced from Columbia. Layer: app — the AI Act regulates uses (Annex III high-risk deployments), with one chapter reaching models (general-purpose AI, obligations live since 2025-08-02).

Key claims (Says)

• Empirical: The Brussels Effect — the EU projects its standards globally because firms adopt EU rules for their whole operation rather than maintain dual standards (glossed: one rulebook is cheaper than two). Held on mechanism / Drifted on reach. Mechanism: of 2,973 firms surveyed, 47% of those citing the AI Act in governance disclosures sit outside the EU, US firms alone 35% (Thomson Reuters Foundation data via Euronews, 2026-07-30); 21 developers including OpenAI, Google, Anthropic, Microsoft and Amazon signed the voluntary GPAI Code (EC list). Reach: she now says the high-water mark is past (Capgemini, 2026-03), the EU cut its own AI Act (Omnibus in force 2026-07-27), and only 13% of surveyed firms have any AI-governance framework at all — the effect is real, concentrated, and shallow.
• Empirical: The AI Act will have Brussels Effect dynamics — risk-tiered classification becomes the de facto global standard even where no equivalent law exists. Drifted. South Korea's AI Basic Act (in force 2026-01-22) is the only other comprehensive national AI law and is explicitly lighter, with fines deferred a year (Cooley); Japan's AI Promotion Act (passed 2025-05-28) carries no monetary penalties by design (FPF); India issued non-binding guidelines instead of a law (MeitY, 2025-11-05); Brazil's risk-based PL 2338 passed the Senate 2024-12-10 and has sat in the Chamber since (Library of Congress); the US moved the other way — an Executive Order to preempt state AI laws (2025-12-11); Colorado's EU-style law was stayed by a federal magistrate (2026-04-27) in a suit the DOJ joined, then narrowed and pushed to 2027-01-01 (2026-05-14). Firm level: Meta refused the GPAI Code (2025-07-18); xAI signed the safety chapter only (2025-07-31). Dual-compliance and market-specific models — v1's own listed weak spot — are what emerged.
• Empirical: Three competing "digital empires" (US market-driven, EU rights-driven, China state-driven) are rival visions for governing technology. Held — the US–EU collision is now open policy: Trump threatened tariffs on any country with "Digital Taxes, Legislation, Rules, or Regulations" (2025-08-25; the Commission replied that no changes to the DSA/DMA "were on the table," Eunews); the Commission's first DSA fine (X, €120M, 2025-12-05) was answered within three weeks by US visa bans on ex-Commissioner Breton and four Europeans (2025-12-23). Author-issued caveat: "everyone is playing Beijing's game now" (Capgemini, 2026-03) — the market-driven empire now uses state muscle (tariffs, DOJ suits, an EO) to defend its firms. The taxonomy holds; the labels blur.
• Empirical: Regulatory export is structural, not political — it runs through market mechanics (firms choosing one standard), not diplomatic pressure. Wrong on the "not political" half. The first real political stress test — Washington tying tariff relief to EU digital rules (CEPA, 2025-12-03), the visa bans, and explicit retaliation threats over DMA/DSA enforcement (Fortune, 2025-12-17) — ran alongside the Omnibus retreat, and she attributes the wavering to "US pressure" herself (Capgemini, 2026-03). Structure explains why rules spread; it did not stop politics from pulling them back. Mechanism half stands; the insulation claim failed.
• Normative: Democratic governance of AI is non-negotiable — systems touching fundamental rights need statutory oversight; a values claim, not an efficiency claim. Still endorsed (Foreign Affairs, 2025-04-21; Capgemini, 2026-03). No observable — unfalsifiable as stated; retained as her normative anchor, not graded.
• Empirical, new (2024; restated 2026-03): Over-regulation is not why Europe lags in tech — fragmentation, capital markets, risk culture and talent are "much bigger factors." Live — check below.
• Empirical, new (2025-04): "Forbearance" — the EU's deeper problem is deliberately under-enforcing its own laws, made worse but not caused by Trump-era pressure (with R. Daniel Kelemen and Tommaso Pavone). Live — check below; the X fine, the first DSA penalty three years after the law took effect, cuts both ways.

Notable predictions — with falsifiable checks

• (2020) The Brussels Effect extends from chemicals and privacy into digital regulation. Held — GDPR became the template, and the AI Act is now cited in governance disclosures by firms on three continents (Euronews/TRF, 2026-07-30). Check: the non-EU share of firms citing EU digital law in the next TRF round (2027); a falling share while the laws stand = the effect fading independent of the Omnibus.
• (2023) Three regulatory models will define AI governance. Held — divergence widened (US EO 2025-12-11 vs the EU high-risk regime). Check: by 2027-12-02, does the US codify a preemptive federal framework while the EU's Annex III obligations actually apply? Convergence in either direction (US adopts risk tiers, or a second EU deferral) = Drifted.
• (Ongoing) The AI Act will be adopted beyond Europe through the Brussels Effect — v1 cited "early evidence in India and other jurisdictions studying the EU model." Drifted; the India evidence is Wrong — India chose no standalone law (MeitY guidelines, 2025-11-05, "lightweight," "techno-legal") and hosted its own AI Impact Summit (2026-02-16 to 21). Check: by 2027-12-02, at least two G20 economies with binding, penalty-backed high-risk obligations recognisably modeled on Annex III (Korea counts as half). Zero = Wrong outright.
• (2026-03, new) "The high-water mark of Europe's regulatory leadership is behind us." Check: through end-2027, no EU digital regulation of DSA/DMA/AI Act scale is proposed AND the 2027-12-02 high-risk date holds. A second deferral makes her right for a worse reason; a new landmark regulation makes her Wrong.
• (2024, new) Deregulation will not close Europe's tech gap. Check: EU AI venture funding and startup formation 2026–28 versus trend after the Omnibus (2026-07-27) — a jump with no capital-market or single-market reform behind it weakens her; flat holds her. Market-level data only, per the Brynjolfsson rule.
• (2025-04, new) Forbearance persists. Check: DSA/DMA non-compliance decisions 2026–27 — fewer than three decisions, none above €500M, by end-2027 = forbearance Held; a run of large fines that survive US pressure = Drifted.

Revealed behavior (Does)

• Time went inside: accepted the Special Advisor seat with EVP Virkkunen (2025), whose portfolio then produced the Digital Omnibus (2025-11-19). Her public verdict on it was measured — "whether you call it 'simplification' or 'deregulation', you are certainly moving away from the high water mark of regulation" (Politico, via Lawfare, 2025-12-15) — and she stayed in the role through the Act's final adoption (Council, 2026-06-29).
• Output: one co-authored Foreign Affairs essay (2025-04-21), one law-review article (2024), two podcasts (Masters of Privacy 2026-02-15; Cleary Antitrust Review 2026-02-19 — linked in Sources), one corporate-research interview (Capgemini, 2026-03). No new book announced. No public statement after 2026-03 was found this cycle, none on the Omnibus's final adoption — an insider's silence, or a scholar between books; undecidable from outside.
• Kept the term: still frames everything as the Brussels Effect while conceding its decline — the brand outlasting the phenomenon is a live risk (Weak spots).
• Institutional seats (Carnegie, Berkman, ECFR and Bruegel profiles, a London Speaker Bureau listing) — an influence stack, not a commercial one; no corporate boards, funds or startup ties found.

Feels

Fear that Europe will trade away the one thing it does distinctively — rights-driven legitimacy — for a deregulation race it cannot win; a wish that "regulation and innovation are compatible" be true, because her life's work rests on it. The sting is that EU leaders themselves adopted the over-regulation story.

Hears

The Virkkunen cabinet and Commission services; the transatlantic legal academy (Kelemen, Pavone); Brussels think tanks (Bruegel, ECFR); Columbia and Harvard. Downstream of regulatory-competition and market-size theory (David Vogel's "California Effect" lineage). Almost nothing from model builders or founders — the technical layer arrives second-hand.

Sees

An inside view of how the Commission actually weighs Washington's pressure against its own rulebook, from a chair in New York — the only bench member positioned in both capitals. Plus fifteen years of comparative data on three regimes, which is why her taxonomy keeps holding when her forecasts drift.

Incentive map

Re-derived this cycle. v1 filed "advisory roles to EU institutions" under track record; it belongs here. She is Special Advisor to the Executive Vice-President whose own package rolled back the AI Act — she cannot say "Virkkunen dismantled the Brussels Effect" while advising her, and the shift from the sharp 2025-04 essay to the measured 2026-03 tone is consistent with that constraint, though the appointment month is unverified and causation is unprovable. Her public brand — two books, a speaking-bureau listing — is the phrase "Brussels Effect," giving her a stake in its continued relevance; credit that she has been the one to call its high-water mark. Columbia salary; no corporate boards, funds or startup interests found; terms of the EC role (paid or not) unverified. Finnish citizen, EU insider by identity. Rule-25 discount: moderate on valence claims about what the EU should do; low on the mechanism scholarship, which predates the appointment. "Unconflicted academic" no longer applies (the Bengio rule).

Theories aligned with

• Regulatory globalization / Brussels Effect theory — her own; the market-size-plus-stringency mechanism
• Rights-based technology governance (the digital-constitutionalism strand)
• Adjacent to Khan's antitrust frame but operating at the international regulatory level rather than domestic enforcement (carried from v1)
• Counterweight to the deregulation strand of techno-optimism — her 2024 "false choice" paper is the bench's most direct rebuttal of "regulation kills innovation"
• Converging, since 2026-03, with state-shapes-markets thinking (Mazzucato): "everyone is playing Beijing's game now"

What she's reacting against

• US-centric framing of AI governance as a US–China bilateral competition — argues the EU is the regulatory third pole
• The assumption that regulation kills innovation — argues the EU model creates predictable rules that enable investment, and that Europe's lag has structural causes
• Tech-industry narratives that self-regulation is sufficient — insists democratic governance of AI requires statutory frameworks
• The view that the EU is merely reactive — argues it actively shapes global market standards through its regulatory ambition
• NEW (2025–26): the EU's own "over-regulation" story and its habit of forbearance; Washington's use of trade and visas to discipline EU rule-making

Where she overlaps / splits (with Rich)

• Overlaps with Schaake (no bench file) on EU governance leadership; splits on register — Bradford the academic theorist, Schaake the practitioner and policy operator. The register gap has narrowed: Bradford is now an operator too.
• Overlaps with Khan on AI needing structural regulation; splits on mechanism — Khan uses antitrust enforcement, Bradford uses regulatory export. Her 2026-02 Cleary podcast on Trump-era antitrust is a data point for Khan's file.
• Overlaps with Mazzucato on the state's legitimate role in shaping technology markets; splits on focus — Mazzucato is industrial policy, Bradford is regulatory standard-setting; the split narrowed in 2026-03.
• Splits with Andreessen and Thiel on regulation's effect on innovation — they see drag, she sees enabling structure. 2025–26 evidence: their side won politically in Brussels (the Omnibus), which vindicates the lobby's power, not its mechanism claim — her NULR paper is still the standing rebuttal.
• Overlaps with Zuboff on rights-driven governance; splits on register — Zuboff diagnoses, Bradford designs regimes.
• Splits with US tech policy broadly on the adequacy of self-regulation — now a formal transatlantic dispute, not a seminar disagreement.
• Data point for Clark: Anthropic signed the GPAI Code in full (EC list); Meta refused and stayed in the market — the code split the labs along the line her theory predicts (market access) and one it does not (refusal without exit).

Track record

• The Brussels Effect (2020) coined and validated a concept now used across policy discourse — a genuine field-defining contribution; even her critics (CEPA 2025-12; Tendil, SSRN 2026) use the phrase as the yardstick for measuring the EU's retreat
• Digital Empires (2023) named an FT best book of 2023; won the 2024 Stein Rokkan Prize (ECPR, verified) — peer recognition is strong
• GDPR as prior evidence: the Brussels Effect thesis was validated by GDPR's global adoption before the AI Act was even drafted
• Advisory roles to EU institutions — influence on actual policy, not just academic commentary; upgraded in 2025 to Special Advisor to an Executive Vice-President, and now also an incentive-map entry (above)
• Elected to the Finnish Academy of Science and Letters, 2026
• Pattern: taxonomy and mechanism calls hold; reach-and-timing forecasts drift; she has been early and candid on the decline of her own concept, which is rare on this bench

Empirical vs normative

• Empirical: how regulatory standards spread across markets — case studies of chemicals (REACH), GDPR, data protection as Brussels Effect precedents; now the AI Act as the live test, with firm-level adoption data
• Normative: democratic governance of AI is a fundamental requirement; the EU model is preferable to market-driven or state-driven alternatives; Europe should enforce, not forbear
• The empirical machinery (Brussels Effect as mechanism) is well-documented and partly confirmed at firm level; the normative preference for the EU model is a values choice and, since 2025, an insider's brief
• Conflict note (CONVENTIONS rule 25): moderate — a Commission advisory seat plus a concept-branded career; no commercial position

Weak spots / open questions

• The Brussels Effect assumes companies prefer one global standard — but AI regulation is complex enough that dual-compliance or market-specific models may emerge. Status 2026-09: emerging — Meta's refusal, xAI's partial signature, Korea's lighter tier.
• EU AI Act enforcement is untested — the regulatory ambition may exceed implementation capacity. Status: partially tested — the Omnibus gave the AI Office exclusive competence over GPAI with inspection and fining powers (Gibson Dunn, 2026-05-27), and the first DSA fine landed (2025-12-05), but the high-risk core was pushed to 2027-12-02 before a single case (EC, in force 2026-07-27).
• The EU model's preference for precaution over speed may cede frontier AI development to the US and China — her frame does not fully address the innovation-cost tradeoff. Status: her NULR answer exists and is contested; the EU itself acted as though regulation were the cost.
• "Rights-driven" framing assumes consensus on which rights matter and how to weight them — contested even within Europe. Almada and Radu's "Brussels Side-Effect" (German Law Journal, 2024-02-19) sharpens this: the AI Act's product-safety framing exports thin rights protection wherever it is copied.
• Less engaged with technical AI questions — the frame is regulatory structure, not model capability or alignment. Unchanged.
• NEW: the Special Advisor seat — what she cannot say, and whether the silence since 2026-03 is chosen or imposed
• NEW: the brand problem — if the phenomenon keeps fading while the phrase persists, read "Brussels Effect" as a historical claim about 2016–2024, not a forecast
• NEW: the GDPR half of the Omnibus is still open — the Cyprus presidency withdrew its text 2026-06-30 for lack of a majority; Ireland resumes in September (Privacy Next). If "legitimate interest" for AI training passes, the flagship of her thesis is the one being rewritten

Rich's take

• (your synthesis here)

Delta log

2026-09-21 — v1→v2 migration + validation (bulk run)

• Review window 2026-06-16 → 2026-09-21; the June review is treated as never run (batch rule). Grades: 4 Held / 3 Drifted / 2 Wrong (1 unfalsifiable normative claim noted, not graded; 3 new live claims given checks).
• Wrong: "regulatory export is structural, not political" — the Washington Pull (tariff threat 2025-08-25; retaliation threats 2025-12-17; visa bans 2025-12-23; all linked in Key claims) ran alongside the Omnibus retreat (EC, in force 2026-07-27), and she credits "US pressure" herself (Capgemini, 2026-03). Wrong: "early evidence in India" — India chose guidelines over law (MeitY, 2025-11-05).
• Drifted: Brussels Effect reach (author-declared high-water mark, 2026-03); AI Act Brussels-Effect dynamics (Korea lighter 2026-01-22; Japan penalty-free; Brazil stalled; US EO 2025-12-11; Meta refusal 2025-07-18 — linked in Key claims); AI Act adoption abroad.
• Held: three digital empires (the X-fine-to-visa-bans sequence, EC 2025-12-05); 2020 digital-extension prediction (TRF firm data, Euronews 2026-07-30); 2023 three-models prediction (US EO vs EU regime); advisory-role track record, upgraded.
• Most surprising delta: she went inside — Special Advisor (2025) to the Commissioner whose Digital Omnibus then cut the AI Act's high-risk deadline by 16 months — and then declared her own concept's high-water mark past.
• Employer/role check (mandatory): Columbia chair unchanged; new role Special Advisor to EVP Virkkunen (Berkman, updated 2026-08-04; Columbia bio); Finnish Academy election 2026. Identity flag: the Jean Monnet Chair claim is Elaine Fahey's, not hers. Incentive map re-derived; "unconflicted academic" retired.
• Post-review sources found: Capgemini 2026-03-04; Omnibus adoption 2026-06-29 / in force 2026-07-27; Euronews/TRF 2026-07-30; Berkman 2026-08-04. No Bradford statement after 2026-03 found; the September 2025 CV PDF could not be text-extracted locally, so the Columbia and Berkman pages stand in for it.
• Tier: semiannual — she publishes about once a quarter, but the GDPR-Omnibus fight (Council resumes 2026-09) and the AI Office's new powers make a March 2027 check worthwhile; wake triggers cover the rest.
• Confidence unchanged at medium — mechanism grades rest on one firm-level dataset; reach grades on her own words plus legislative facts.

Sources

• Her words, 2024–26: Capgemini interview (2026-03; PDF dated 04-03-2026, index re-listed 05-29) · Foreign Affairs (2025-04-21, with Kelemen and Pavone) · Lawfare, quoting her Politico remark (2025-12-15) · "The False Choice…" (NULR 2024) · Masters of Privacy (2026-02-15) · Cleary Antitrust Review (2026-02-19) · Project Syndicate columns
• Role and honors: Columbia faculty page · Berkman Klein profile (2026-08-04) · City St George's Jean Monnet Chair (holder: Fahey) · ECPR — Stein Rokkan Prize 2024
• EU legislation and enforcement (primary): EC — AI Omnibus in force (2026-07-27) · Council adoption (2026-06-29) · Gibson Dunn — Omnibus changes (2026-05-27) · EC — GPAI Code signatories · EC — X fined €120M (2025-12-05) · Privacy Next — GDPR Omnibus (2026-09)
• Adoption elsewhere: Cooley — Korea (2026-01-27) · FPF — Japan · MeitY — India (2025-11-05) · Library of Congress — Brazil (2025-05-23) · Euronews / TRF firm data (2026-07-30)
• US pushback: Gibson Dunn — preemption EO (2025-12-11) · Euronews — tariff threat (2025-08-26) · Eunews — "we decide our own rules" (2025-08-26) · Fortune — DMA/DSA retaliation threat (2025-12-17) · Euronews — visa bans (2025-12-24) · Colorado AI law stayed and rewritten (2026-05) · TechCrunch — Meta refuses the Code (2025-07-18) · xAI signs safety chapter only (2025-07-31)
• Critiques: CEPA (2025-12-03) · Almada & Radu, German Law Journal (2024-02-19) · Tendil, SSRN (2026) · GovAI — "The Brussels Effect and AI" (2022, from v1)
• Books: The Brussels Effect: How the European Union Rules the World (OUP 2020) · Digital Empires: The Global Battle to Regulate Technology (OUP 2023-09)

migrated v1→v2 2026-09-21 · backup: _archives/anu-bradford.html.bak-20260921 · converts-from: people/anu-bradford.md · AI & Society domain