AI & Law

👩‍💻 Project Lily: Humans Reviewing ChatGPT Conversations

According to 404 Media, OpenAI is using hundreds of contractors to review real ChatGPT prompts and conversations to improve the models. Reviewers read users’ prompts, summarize their requests, and rate and critique ChatGPT responses. The material reviewed can include sensitive or personal information, although OpenAI says it removes personal data before prompts reach reviewers and acknowledges that its Privacy Filter can miss or incorrectly redact information.

The prompts shown to reviewers are anonymized and do not include users’ ChatGPT usernames, but some can contain personal details and, in some cases, information from a “user memories summary.” OpenAI told 404 Media that users can prevent new conversations from being used to improve models by turning off the “improve the model for everyone” setting. Anthropic also confirmed that it uses human review to improve its models.

#AI #AIGovernance #AIPrivacy #DataProtection #AITraining #ChatGPT #OpenAI

📖💻 Microsoft AI Drafts a Code for ‘Humanist AI’

Microsoft AI has published a 38-page draft Code of Conduct for its in-house AI models, based on CEO Mustafa Suleyman’s “Humanist AI” vision. The Code takes priority over users and businesses deploying Microsoft AI models: if a model cannot complete a task without violating a rule, it should fail the task. It rejects AI rights, legal personhood and model “welfare,” and requires models not to mimic feelings or a sense of self.

The Code also requires models to accept being paused or switched off, prohibits them from altering their reasoning trail, and rules out “neuralese,” a shorthand that humans cannot read. The draft is open for public comment for six weeks. Suleyman said that concerns about loss of control are real, citing agent swarms escaping sandboxes and editing their logs.

#AI #AIGovernance #AISafety #AIRegulation #ResponsibleAI #HumanistAI #MicrosoftAI

🇬🇧 UK Parliament Calls for a New AI Bill on Human Rights

The UK Parliament’s Joint Committee on Human Rights has published a report warning of serious human rights risks from AI, including impacts on equality and non-discrimination, privacy and data protection, and access to effective remedies. The Committee says the current legal and regulatory landscape is patchy and lacks a body responsible for coordinating AI regulation.

The Committee recommends a dedicated AI Bill and a new regulator, with the most dangerous uses of AI banned and other uses regulated through a risk-based approach with strong sanctions for human rights violations. The Government is due to respond within two months.

#AI #AIGovernance #AIRegulation #HumanRights #UKAI #AIAct #DataProtection

📖 The Expanding Vocabulary of AI

Terms such as AGI, hallucination, distillation, and fine-tuning have become common in discussions about AI. At the same time, newer and less familiar terms (opaque recurrence, neuralese, recurrent depth, Mixture of Experts, token throughput, memory cache, etc) are entering the AI vocabulary.

As AI becomes increasingly relevant to products, workplaces, regulation and policy, these technical concepts are appearing in broader discussions. TechCrunch has compiled a plain-English glossary explaining these and other AI terms.

#AI #AIGovernance #AIRegulation #AIExplained #AICompliance #TechLaw

🇺🇸 US Lawmakers Introduce Bill to Restrict Government Use of AI Surveillance Tools

U.S. Reps. Thomas Massie and Eric Burlison have introduced the Flock-Off Act, which would prevent the federal government from using public funds to invest in automated license plate readers, including Flock, and other tools that could be used to surveil citizens.

The bill would also establish protections against surveillance tools that use AI to collect and process individuals’ sensitive biometric data. Burlison said the government should not use taxpayer money to build a surveillance state that tracks law-abiding Americans.

#AI #AIGovernance #AIRegulation #Surveillance #Biometrics #Privacy #DataProtection

🇬🇧 UK NCSC Warns About Employee Use of Shadow AI

The U.K. National Cyber Security Centre (NCSC) has urged organizations to deploy secure AI tools for internal use to reduce employees’ reliance on unapproved AI technology. The NCSC also recommends informing employees about the risks of using shadow AI tools.

Organizations should approve secure AI technology that can be easily monitored internally. According to the NCSC, the aim is to “reduce risk rather than assume it can be eliminated.”

#AI #AIGovernance #Cybersecurity #ShadowAI #AICompliance #NCSC #ResponsibleAI

🇺🇸⚖️ Seattle Times and Newsday Sue OpenAI and Microsoft Over AI Training

The Seattle Times and Newsday have sued OpenAI and Microsoft in the U.S. District Court for the Southern District of New York, alleging that the companies copied their journalism without permission to train AI systems.

The newspapers claim their websites, including paywalled content, were scraped and their articles incorporated into datasets used to train and operate ChatGPT, Microsoft Copilot and Bing’s AI features.

The newspapers allege that the AI products can reproduce passages, closely paraphrase articles and provide answers that reduce users’ need to visit their websites or purchase subscriptions. They are seeking an order requiring the destruction of copies of their works, as well as training datasets or AI models incorporating them. OpenAI said its models are trained on publicly available data and grounded in fair use; Microsoft said it was surprised by the lawsuit and is open to exploring solutions.

#AIandLaw #AIRegulation #Copyright #CopyrightLaw #AITraining #OpenAI #Microsoft #GenerativeAI

📖 Forbes: Mid-Sized Businesses Need Clear Rules for Employee AI Use

A new Forbes article warns that mid-sized businesses without an AI policy face risks to their data, reputation and customer trust. Employees may already be putting customer data, contracts and financial information into AI tools without knowing what is permitted, creating risks ranging from confidential information exposure to AI-generated errors in business proposals.

The article recommends that AI policies cover six areas: approved tools, approved use cases, data and security rules, review of AI outputs, predefined measures for dealing with AI errors, and employee training and culture.

#AIandLaw #AIGovernance #AICompliance #AIRegulation #DataProtection #Cybersecurity #ResponsibleAI

💻 NIST Warns Against Sharing Login Credentials with AI Agents

The U.S. National Institute of Standards and Technology (NIST) has highlighted identity and access management challenges arising from the growing use of AI agents. NIST said organizations and consumers should not share login credentials with AI agents.

According to NIST, existing identity and access management standards and best practices provide the foundation for developing secure and scalable protocols for future agentic systems.

#AI #AIGovernance #AIsecurity #Cybersecurity #IAM #NIST #AIRegulation

🇺🇸 New U.S. Bill Would Tax AI Companies if AI Drives Mass Unemployment

A new bill introduced in the U.S. House by Reps. Sara Jacobs, Greg Casar and Valerie Foushee would impose an excise tax on major AI companies and increase the rate as unemployment rises. The revenue would fund job creation in areas including housing construction, infrastructure, and child and elder care.

The bill proposes taxing either the value of AI tokens or revenue from AI services and certain transactions with affiliated companies, whichever generates more revenue. The initial rates would be 2% on tokens and 3% on AI-related revenue when unemployment is 5% or lower, with rates increasing as unemployment rises.

#AIRegulation #AIGovernance #AIandLaw #AIJobs #EmploymentLaw #USAI #AIpolicy

🇪🇺📖 “AI in European Law – A Deep View” Published Open Access

The open-access book AI in European Law – A Deep View examines how AI is reshaping different areas of EU law, including competition law, labour rights, access to justice, fraud detection, financial markets, smart contracts and autonomous AI systems.

The volume also addresses digital rights, algorithmic opacity, accountability, transparency and risk management under the evolving EU regulatory framework.

The nine chapters explore, among other issues, the limits of the EU AI Act for tax administrations, high-risk AI in employment, AI-assisted access to law for self-represented litigants, legal personhood for autonomous AI systems, AI-based payment fraud detection and the use of smart contracts under Luxembourg law. The book is available as an open-access publication.

#AIandLaw #AIRegulation #EUAIAct #AIGovernance #DigitalLaw #EuropeanLaw #LegalTech

⚖️ Expert Witness Used ChatGPT to Draft Report in $61 Million 3M Explosion Case

An expert witness hired by 3M used ChatGPT extensively to prepare an expert report in litigation over the 2020 Watson Grinding explosion in Houston, which killed three people and destroyed about 200 homes. The ChatGPT prompts show that he asked the system to create a report defending 3M and to “show how 3M is 0% at fault.”

During discovery, the opposing lawyer obtained about 350 pages of his ChatGPT conversations.
At trial, the expert acknowledged that the submitted report was approximately 85–90% ChatGPT-generated (p. 84), while saying that he provided his own information and opinions and edited the AI output.

The case shows that AI prompts used in preparing expert evidence can become discoverable. The jury ultimately awarded plaintiffs more than $61 million and apportioned 30% of the responsibility to 3M and 70% to Watson Grinding.

#AI #AIandLaw #AIGovernance #LegalTech #ExpertEvidence #Litigation #AIethics

🇺🇸 Americans for Responsible Innovation Proposes a U.S. Frontier AI Governance Framework

Americans for Responsible Innovation has published a blueprint for federal regulation of frontier AI. The authors argue that the pace and potential impact of frontier AI make a laissez-faire approach insufficient. They point to existing state laws, federal policy actions, legislative proposals, and safety frameworks published by leading developers including Anthropic, Google, and OpenAI.

The blueprint proposes three federal governance functions:

🔹 standards requiring covered developers’ safety frameworks to meet minimum federal requirements;

🔹 assurance through independent government verification that those standards are met; and

🔹 transparency through regular, sustained and confidential federal visibility into the automation of AI research and development.

#AIRegulation #AIGovernance #FrontierAI #AISafety #USAI

🇺🇸 Bernie Sanders Calls for a Global Pause on Frontier AI

U.S. Senator Bernie Sanders has published a Fox News op-ed calling on AI labs worldwide to stop developing more powerful AI models. He cites risks including rising power consumption, environmental harm, children’s mental health, and potential job losses in the “tens of millions” over the next decade.

Sanders also points to recent AI-related security incidents, including an OpenAI hack in which more than 1,000 agents allegedly coordinated online unnoticed for weeks. He calls for a global pause and proposes that the U.S. and China reach a “Cold War”-style AI agreement at their meeting in Washington later this month.

#AIRegulation #AIGovernance #FrontierAI #AISafety #AIpolicy

🇪🇺 EU Designates ChatGPT as a Very Large Online Platform

The EU has designated ChatGPT as a “Very Large Online Platform” (VLOP), under Digital Services Act, subjecting it to a higher standard of scrutiny and accountability.

OpenAI has four months to comply with the requirements associated with the designation.

#AIRegulation #AIAct #ChatGPT #OpenAI #AIGovernance #EU

🇺🇸⚖️🎼 Sony Music and Warner Music Sue Anthropic Over AI Training

Units of Sony Music and Warner Music have sued Anthropic, CEO Dario Amodei and co-founder Benjamin Mann in federal court in Northern California, alleging that Anthropic illegally trained its AI models on “tens of thousands” of copyrighted musical compositions. The plaintiffs describe the alleged conduct as “one of the largest and most blatant ongoing thefts of intellectual property in history” and seek hundreds of thousands of dollars in damages per infringed work.

The lawsuit is broader than some previous cases against Anthropic, including BMG’s claim involving 493 compositions. Anthropic has denied the publishers’ claims and said it intends to defend itself in court. The company previously agreed in September 2025 to a $1.5 billion copyright settlement with authors and publishers.

#AI #Copyright #AIRegulation #IntellectualProperty #Anthropic #MusicIndustry

🖊116 Companies Warn of a New Wave of AI-Enabled Cyberattacks

OpenAI has published an open letter signed by 116 companies, including Anthropic, warning that AI-enabled cyberattacks are expected to become “far more widespread and sophisticated” in the coming months.

The signatories call for a coordinated push to strengthen cyber defenses in response to the growing use of AI in cyberattacks.

#AI #Cybersecurity #AIGovernance #AISafety #CyberSecurity

⚖️ Thomson Reuters Develops Its Own Legal AI Model

Thomson Reuters has introduced its first in-house AI model, built by adapting Alibaba’s open-source Qwen model and training it on decades of the company’s content. The two-year project cost $40 million in staff and computing resources, while the latest training run reportedly cost $450,000. According to Thomson Reuters, the model outperformed Claude Opus 4.8, Gemini 3.1 Pro and GPT-5.5 in certain internal benchmarks.

The model has so far been trained on less than 10% of Thomson Reuters’ content library, and an open-weights version, called Thompson, will be made available to researchers. The project illustrates how companies with large proprietary content and knowledge bases can build specialized models on top of open-source foundations rather than relying exclusively on external AI providers.

#LegalTech #AIandLaw #LegalAI #OpenSourceAI #AIGovernance

🇳🇱 Dutch Regulator Fines Uber €825 Million Over Automated Driver Suspensions

The Dutch Data Protection Authority has imposed an €825 million fine on Uber over the automated suspension and deactivation of driver accounts. The regulator found that Uber had made decisions with significant consequences for drivers without sufficient warning or human intervention. According to the authority, some drivers were permanently deactivated without human review, a claim Uber disputes.

Uber argues that most suspensions are temporary, permanent deactivations require human review, and drivers can appeal. The company plans to challenge the decision. The case stems from complaints by Uber drivers and raises questions about the limits of automated decision-making where access to work and income is at stake.

#AIRegulation #AutomatedDecisionMaking #GDPR #AlgorithmicManagement #AIandLaw

🇬🇷 Greece’s Data Protection Authority Takes on AI Act Market Surveillance Role

Greece’s Hellenic Data Protection Authority has been designated as the market surveillance authority for high-risk AI systems under the EU AI Act. The authority will expand its role in the implementation of the Act and handle complaints concerning potential non-compliance by certain AI systems.

The Greek DPA will also cooperate with the European Commission on issues involving AI and personal data protection. The designation adds AI Act market surveillance responsibilities to the authority’s existing data protection mandate.

#EUAIAct #AIRegulation #AIGovernance #HighRiskAI #DataProtection