Tech & Law Digest

What if a legal prediction model looks accurate only because it is reading hindsight baked into court decisions?This explainer walks through "Shortcut Learning in Legal Judgment Prediction: Empirical Evidence from the UK Employment Tribunal", focusing on why post-hoc judicial texts can make outcome prediction look stronger than it really is, how shortcut learning can enter legal AI systems, and what this means for evaluating legal judgment prediction responsibly.Source: Shortcut Learning in Legal Judgment Prediction: Empirical Evidence from the UK Employment TribunalAuthors: Joe Watson, Joana Ribeiro de Faria, Marcus Tomalin, Måns Magnusson, Huiyuan Xie, Hao Tian Yeung, Christine Carter, Jonathan Rutherford, Felix SteffekLink: https://arxiv.org/abs/2607.04261v1Attribution: This video is an educational digest/commentary based on the source above. All credit for the original research belongs to the authors. Any simplification, framing, or visual explanation is for study and public understanding.Produced as a NotebookLM Video Overview and reviewed before publication.This is for educational and research discussion only, not legal advice.Hooks / key questions:- Can legal AI predict outcomes, or is it learning from hindsight in judicial reasons?- Why can high accuracy hide a weak forecasting setup?- What should evaluation look like when legal prediction models use post-hoc legal texts?#LegalAI #LawAndTechnology #TechLawDigest

What if a court-facing LLM refuses a lawful translation or summary task simply because the case facts are disturbing?This video explains "Measuring & Mitigating Over-Alignment for LLMs in Multilingual Criminal Law Courts" by Arthur Wuhrmann, Gaetan Stein, Daniel Brunner, and Andrei Kucharavy.The paper studies how multilingual LLMs behave on criminal-law court tasks where the material may be graphic, sensitive, or emotionally difficult while still being lawful and professionally necessary.Main points covered:- What over-alignment means in legal AI workflows, and why it differs from ordinary safety refusal- Why multilingual criminal-law tasks create a hard test for aligned LLMs- How the authors build a benchmark around lawful court tasks that models often refuse- What the results show about refusal behavior, language effects, and model family differences- Why prompt engineering alone does not fully solve the problem- How mitigation methods including abliteration and model choice affect court-usable performance- What this means for legal tech teams deploying LLMs in high-stakes multilingual settingsPaper:arXiv: Measuring & Mitigating Over-Alignment for LLMs in Multilingual Criminal Law Courtshttps://arxiv.org/abs/2606.23375This content may discuss criminal case materials, including sensitive or graphic factual scenarios, strictly for research and educational analysis.This content is provided for informational purposes only and does not constitute legal advice. You are responsible for how you use this information and should seek qualified advice for specific matters.#LegalAI #LLMSafety #AIGovernance #CriminalLaw #LegalTech #MultilingualAI #AIAlignment #CourtTechnology

Are legal AI models learning the law, or just learning the judge?This video explains "Towards Explainable Adjudicative Variance: Quantifying Judicial Discretion via Gated Multi-Task Learning" by Stanisław Sójka, Felix Steffek, and Matthias Grabmair.The paper studies legal outcome prediction on 13,937 UK Employment Tribunal decisions and asks whether models can separate objective case facts from adjudicative context and judge-specific discretion.Main points covered:- Why judicial discretion matters when evaluating legal NLP systems- How a judge-aware gated multi-task learning architecture models shared legal structure alongside judge-level variance- Why the paper introduces a fine-grained outcome taxonomy to regularize the encoder- How the authors compare their architecture against prompt-based supervised fine-tuning baselines- Why the gains matter most for ambiguous and rare outcome classes- What interpretable judge embeddings and calibration profiles reveal about adjudicative context- How legal AI teams should think about prediction, explanation, and institutional use in court-related settingsPaper:Sójka, Steffek, and Grabmair, "Towards Explainable Adjudicative Variance: Quantifying Judicial Discretion via Gated Multi-Task Learning"https://arxiv.org/abs/2606.27069This content is provided for informational and educational purposes only and does not constitute legal advice.#LegalAI #LegalNLP #JudicialDiscretion #ExplainableAI #MachineLearning #CourtTechnology #AIResearch

What happens when a stablecoin becomes important enough to be treated like financial infrastructure?This video explains the Bank of England's June 2026 policy statement and consultation on sterling-denominated systemic stablecoins.The framework matters because it shows how the UK plans to let sterling stablecoins scale while protecting trust in money, payments, credit provision, and financial stability.Main points covered:- Why systemic stablecoins sit between payments innovation and financial stability regulation- How the Bank of England and FCA roles fit together in the UK stablecoin regime- The shift from the earlier 60/40 reserve proposal to a 70/30 backing asset framework- Why short-term UK government debt and Bank of England deposits matter for redemption confidence- The replacement of individual holding limits with a temporary GBP 40 billion issuance guardrail- What the draft Code of Practice means for issuers, safeguards, liquidity, and wind-down planning- Why the Bank is trying to balance market entry, competition, and systemic risk controlsSource:Bank of England, "Sterling-denominated systemic stablecoins: Policy statement and consultation on draft Code of Practice," published 22 June 2026.https://www.bankofengland.co.uk/paper/2026/ps/sterling-denominated-systemic-stablecoinThis content is provided for research and educational purposes only. It is not legal, financial, or investment advice.#Stablecoins #BankOfEngland #FinTech #DigitalMoney #Payments #FinancialStability #CryptoRegulation #UKFinTech #CentralBanking

What if an LLM's answer could be traced back to the training examples that most shaped it?This video summarizes "Data Attribution in Large Language Models via Bidirectional Gradient Optimization" by Frédéric Berdoz, Luca A. Lanzendörfer, Kaan Bayraktar, and Roger Wattenhofer.The paper introduces DABGO: Data Attribution via Bidirectional Gradient Optimization. Instead of only asking how training data affects a model output, DABGO flips the question: how would the training data be affected if the generated output were optimized back into the model?Main points covered:- Why training data attribution matters for AI governance, accountability, debugging, and provenance- The difference between forward influence and backward influence- How DABGO uses both gradient descent and gradient ascent on generated text- Why bidirectional loss changes can reveal influential training samples- How the method handles open-ended text generation instead of only single-token fact tracing- What the Wikipedia and Gutenberg experiments show about factual and stylistic attribution- How DABGO compares with BM25, TrackStar, and GeckoSource:Frédéric Berdoz, Luca A. Lanzendörfer, Kaan Bayraktar, and Roger Wattenhofer, "Data Attribution in Large Language Models via Bidirectional Gradient Optimization," arXiv:2606.04928, June 3, 2026.https://arxiv.org/abs/2606.04928Code:https://github.com/ETH-DISCO/DABGOThis content is provided for research and educational purposes only.#LLM #AIInterpretability #TrainingData #DataAttribution #DABGO #AIGovernance #MachineLearning #LLMs

Can AI do doctrinal legal reasoning, or are current benchmarks measuring the wrong thing?This video summarizes Michèle Finck's article "The Measurement Gap in the Automation of EU Law: Benchmarking Doctrinal Legal Reasoning under the EU AI Act."The central argument is that legal AI evaluation has a measurement gap. Many benchmarks test retrieval, classification, extraction, exam-style answers, or other paralegal-adjacent tasks. But those tests do not show whether a model can perform doctrinal legal reasoning: the interpretive work of synthesizing legal sources, producing a defensible account of what a norm requires, and fitting that account into the wider legal system.Main points covered:- Why doctrinal legal reasoning is different from ordinary legal text generation- Why current legal AI benchmarks miss the core interpretive task- How EU law intensifies doctrinal reasoning through internalism, normativity, contestability, and coherence- Why the EU AI Act's accuracy requirement for high-risk judicial AI creates a legal measurement problem- Why Article 15 makes benchmarking more than a technical question- What failure modes a real EU-law doctrinal reasoning benchmark would need to detectSource:Michèle Finck, "The Measurement Gap in the Automation of EU Law: Benchmarking Doctrinal Legal Reasoning under the EU AI Act" (2026).This content is provided for research and educational purposes only and does not constitute legal advice.#EUAIAct #AILaw #LegalAI #LegalReasoning #LegalTech #LLM #Benchmarking

What if most stablecoin transfers are not simple payments at all?This video summarizes the BIS Working Paper "The anatomy of stablecoin transactions" by Fabian Schaer, Anneke Kosse, Tara Rice, Takeshi Shirakami, and Jirapat Siridhasanakul.The paper studies USDT, USDC, and PYUSD activity on Ethereum using 593 million event logs from 141 million transactions. Its core point is simple but important: a stablecoin transfer is often only one event inside a larger on-chain transaction.Main points covered:- Why stablecoin transfers should not automatically be treated as standalone payments- The difference between a transfer event and the full transaction around it- How atomic execution bundles swaps, lending, arbitrage, liquidity provision, and settlement- Why 31.6% of stablecoin transactions are complex- Why 59.96% of transfer events occur inside complex transactions- How USDT, USDC, and PYUSD differ in transaction structure, urgency, co-usage, and timing- Why this matters for stablecoin measurement, market monitoring, policy, and regulationSource:Fabian Schaer, Anneke Kosse, Tara Rice, Takeshi Shirakami, and Jirapat Siridhasanakul, "The anatomy of stablecoin transactions," BIS Working Papers No 1359, Bank for International Settlements, June 2026.https://www.bis.org/publ/work1359.htmThis content is provided for research and educational purposes only and does not constitute legal, financial, regulatory, or investment advice.#BIS #Stablecoins #USDC #USDT #PYUSD #Blockchain #DeFi #CryptoResearch

This video summarizes the arXiv paper "Tracing Stablecoin Contagion during the USDC Depeg after the Silicon Valley Bank Collapse" by Krongtum Sankaewtong, Stefan Kitzler, Bernhard Haslhofer, and Yuichi Ikeda.The paper studies the March 2023 USDC depeg after the collapse of Silicon Valley Bank, using Ethereum ERC-20 transaction data to trace how stress moved through stablecoins and related crypto liquidity assets.Main points covered:

• Why the SVB collapse created a shock to USDC
• How stablecoin transaction activity synchronized during the crisis
• Why USDC-related assets showed broad user activation
• How USDT, WBTC, and WETH absorbed stress through larger value flows
• How accounts shifted from single-asset to multi-asset positioning
• Why USDT became a short-lag liquidity destination
• What intraday activity rhythms and balance-size differences reveal
• Why stablecoin contagion should be monitored as a behavioral network process, not just a price event

• Krongtum Sankaewtong, Kyoto University
• Stefan Kitzler, Complexity Science Hub and AIT Austrian Institute of Technology
• Bernhard Haslhofer, Complexity Science Hub
• Yuichi Ikeda, Kyoto University and Nagoya City University

Authors:Source:Sankaewtong, Kitzler, Haslhofer, and Ikeda, "Tracing Stablecoin Contagion during the USDC Depeg after the Silicon Valley Bank Collapse," arXiv:2606.07442v1, submitted June 5, 2026.https://arxiv.org/abs/2606.07442v1This content is provided for research and educational purposes only and does not constitute legal, financial, regulatory, or investment advice.#USDC #Stablecoins #DeFi #Blockchain #SVB #CryptoResearch

BIS Unified Ledger Explained | Tokenisation, Wholesale CBDC, and Future Money

What happens when central bank money, tokenised deposits, and real-world assets all live on the same programmable platform?

In this video, we break down the BIS Annual Economic Report 2023 chapter "Blueprint for the future monetary system: improving the old, enabling the new" and explain the core architecture behind the unified ledger.

Topics covered:

- What tokenisation actually means

- Why the BIS treats tokens as executable objects

- How the "ramp" connects traditional databases to programmable platforms

- Why wholesale CBDC acts as the settlement anchor

- Why the BIS prefers tokenised deposits over stablecoins

- How the unified ledger is structured

- Atomic settlement, DvP, PvP, and trade finance use cases

- The legal, technical, governance, and privacy challenges

Timestamps

00:00 Introduction and disclaimer

00:30 Why the BIS blueprint matters

01:30 Tokenisation and executable objects

02:59 The ramp between legacy assets and programmable platforms

04:05 Settlement finality and wholesale CBDC

05:13 Tokenised deposits vs. stablecoins

05:59 Why tokenised deposits preserve the two-tier monetary system

07:37 The unified ledger architecture

09:04 Ledger partitions and confidentiality

09:16 Atomic settlement

09:51 DvP and PvP settlement

10:41 Trade finance on a unified ledger

11:04 The four implementation challenges

12:12 Final synthesis

12:42 Closing

Source

Bank for International Settlements, Annual Economic Report 2023, Chapter III:

"Blueprint for the future monetary system: improving the old, enabling the new"

This content is provided for research and educational purposes only and does not constitute legal, financial, regulatory, or investment advice. You are responsible for how you use this information and should seek qualified professional advice for specific matters.

#BIS #CBDC #Tokenisation

When a Bot Signs a Contract, Who's Bound? Chopra & White Explained

This content is provided for informational purposes only and does not constitute legal advice. You are responsible for how you use this information and should seek qualified advice for specific matters.

In this video, we break down Samir Chopra and Laurence White's classic analysis of autonomous contracting and the "contracting problem" created by artificial agents.

The paper asks a foundational question for digital commerce: when a bot forms a contract, who should be legally bound by it? Rather than inventing a brand-new legal category for AI systems, Chopra and White argue that agency law already offers a workable answer.

The paper walks through:

- why the "mere tool" theory starts to fail as software becomes more autonomous

- how actual authority and apparent authority can be translated into algorithmic settings

- the paper's taxonomy of specification errors, induction errors, and malfunctions

- why agency law can allocate risk more coherently across operators, users, and third parties

- how existing legislative approaches compare to the agency-law model

Source paper: Samir Chopra & Laurence White, 'Artificial Agents and the Contracting Problem: A Solution Via an Agency Analysis' (2009) University of Illinois Journal of Law, Technology & Policy 363-380.

If you're interested in AI contracts, autonomous agents, legal tech, and the future of digital commerce, this episode gives you a compact map of one of the field's most important early arguments.

#AIContracts #LegalTech #AgencyLaw #AutonomousCommerce #ContractLaw