Dazza Greenwood

The DABUS Post-Mortem: Why the Supreme Court Just Hard-Coded "Human Authorship" ⚖️💻 The U.S. Supreme Court has declined to review the petition from Stephen Thaler, effectively upholding the USPTO and lower court stances: AI-generated outputs are ineligible for copyright protection. For the engineering and R&D community, this isn't just a "legal quirk"—it’s a major architectural constraint for AI-integrated product lifecycles. The Technical & Legal Infrastructure Gap The refusal to grant copyright to Thaler’s DABUS system confirms a binary in IP law: Human as Architect: AI can be a tool in the stack (like a compiler or an IDE), but the "creative spark" must be human-driven. Autonomous Output = Public Domain: If the "expression" is generated via a black-box prompt where the human lacks granular control over the final bits, that output is technically born into the public domain. Key Implications for Tech Leaders: Asset Devaluation: If your core product relies on generative outputs (synthetic data, auto-generated code, or UI assets), those assets may have zero legal defensibility against competitors. The "Prompt" vs. "Control" Problem: The Copyright Office does not view a text prompt as "authorship." High-level orchestration is not the same as creative execution in the eyes of the law. Compliance & Chain of Title: Companies must now document the "human-in-the-loop" contribution to prove that AI was an instrument, not the author. Without this, your IP portfolio is at risk of being unregistrable. The Bottom Line We are entering an era where the most valuable technical systems will be those that facilitate collaborative creativity—where AI augments the human rather than replacing them. If your business model assumes ownership of 100% AI-generated content, it’s time to refactor your IP strategy. How is your team documenting human intervention in AI-assisted workflows? #SoftwareEngineering #DataScience #IPLaw #GenAI #MachineLearning #TechStrategy #OReilly

1

𝗖𝗮𝗹𝗶𝗳𝗼𝗿𝗻𝗶𝗮 𝗣𝗮𝘀𝘀𝗲𝘀 𝗙𝗶𝗿𝘀𝘁-𝗼𝗳-𝗜𝘁𝘀-𝗞𝗶𝗻𝗱 𝗗𝗲𝘃𝗶𝗰𝗲-𝗟𝗲𝘃𝗲𝗹 𝗔𝗴𝗲 𝗩𝗲𝗿𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻 𝗟𝗮𝘄. 𝗛𝗲𝗿𝗲'𝘀 𝗪𝗵𝗮𝘁 𝗘𝗱𝗧𝗲𝗰𝗵 𝗮𝗻𝗱 𝗞𝗶𝗱𝗧𝗲𝗰𝗵 𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗿𝘀 𝗡𝗲𝗲𝗱 𝘁𝗼 𝗞𝗻𝗼𝘄 Governor Newsom signed AB 1043 this week, and it's a game-changer for how platforms handle minor user data. Unlike similar laws in Texas and Utah that were quickly challenged in court, California worked directly with tech companies to design a constitutionally defensible framework. The key differences:  • Verification at the OS level (Apple, Google, Microsoft), NOT the app store.  • Covers ALL computing devices, not just mobile.  • No parental consent required for downloads.  • No photo ID uploads, avoiding the privacy backlash that plagued the UK's Online Safety Act.  • Creates "deemed knowledge" so developers can't claim they didn't know users were minors. The 𝗹𝗮𝘄 𝘁𝗮𝗸𝗲𝘀 𝗲𝗳𝗳𝗲𝗰𝘁 𝗝𝗮𝗻𝘂𝗮𝗿𝘆 𝟭, 𝟮𝟬𝟮𝟳, with operating systems required to be 𝗳𝘂𝗹𝗹𝘆 𝗰𝗼𝗺𝗽𝗹𝗶𝗮𝗻𝘁 𝗯𝘆 𝗝𝘂𝗹𝘆 𝟭, 𝟮𝟬𝟮𝟳. Tech giants backed it. Hollywood opposed it. If you're building apps for kids, this law fundamentally changes your compliance obligations. I've broken down what it means, how it differs from Texas/Utah laws, and what you need to do to prepare. https://lnkd.in/grNBjRHd

24

1 Comment

While No One Was Watching, New York Regulated Frontier AI While much of the AI debate still frames the U.S. as a leader because it avoids regulation, a major development quietly tells a different story. On December 19, 2025, New York Governor Kathy Hochul signed the Responsible AI Safety and Education (RAISE) Act into law. The law will take effect in January 2027. It is one of the most comprehensive state-level governance frameworks for frontier AI models enacted to date imposing binding obligations on large developers of advanced (“frontier”) AI models developed, operating in or affecting New York. It adopts a risk-based, compute-threshold approach similar to California’s Transparency in Frontier Artificial Intelligence Act (TFAIA), while going further in several key areas. The law applies to large developers of “frontier models”, defined by: • Training runs exceeding 10²⁶ computational operations and $100M+ in compute costs, or • Distilled models derived from such systems with $5M+ compute costs. Under RAISE, large developers may not deploy a frontier model if it poses an unreasonable risk of “critical harm.” They must, among other things: • Put safety and security protocols in place before deployment • Assess and mitigate risks such as mass-casualty events, CBRN weapons, autonomous crime, and loss of control • Conduct documented testing and annual reviews • Publish redacted safety plans and submit them to regulators • Report safety incidents within 72 hours • Keep records for the model’s full lifecycle plus five years “Critical harm” is defined at a very high threshold (e.g., 100+ deaths or $1B in damage), while the meaning of “unreasonable risk” will likely be clarified through regulation and litigation. Overall, RAISE closely mirrors California’s approach, but diverges on scope, disclosures, timelines, and penalties, signaling an emerging though still fragmented U.S. framework for governing frontier AI. https://lnkd.in/dwjgbWNi #RAISEAct #responsibleIA

12

3 Comments

Use of Large Language Models in Standards Work #webStandards #ai #LLM "In some cases, there are serious risks. First, there is a possibility of copyright infringement. As LLMs may be trained on copyrighted materials, even well-meaning contributions could create associated liability. There could also be security risks. Providing Member Confidential information to LLMs could result in data leakage (we suggest particular caution with free products). Sometimes there’s subtle incorrectness. LLM-generated text and code may contain subtle falsehoods, which are hard to spot and difficult to debug." https://lnkd.in/e7zErq2n

6

1 Comment

From what I’ve been seeing, affirmative AI coverage basically will require understanding how AI is governed, deployed, and controlled over time within the entity. Some thoughts on issues to consider: Governance and accountability AI-related losses may stem from unclear ownership and fragmented oversight. • Clear enterprise-wide ownership of AI risk • Defined approval, oversight, and review processes • Documented AI policies aligned with risk appetite • Active involvement from senior leadership and the board Use cases Risk arises from what AI is allowed to do and where it sits in the business. • Specific, documented AI applications • Decision-support tools instead of systems that drive outcomes • Functions where AI materially affects customers, employees, or revenue Model development and sourcing Understanding how models are built and maintained will be key. • In-house models versus third-party or open-source tools • Vendor diligence, audit rights, and contractual risk allocation • Controls around retraining, updates, and version management Data governance Data decisions may impact both performance and liability. • Sources and permissions for training and operational data • Use of personal, sensitive, or regulated data • Testing, validation, and documentation of bias and performance HITL Human involvement will be central to defensibility and claims analysis. • Defined points for human review and override • Escalation paths for unexpected or high-impact outputs • Guardrails for use in sensitive or regulated contexts Security and resilience AI may dramatically expand the digital footprint and correlated risks. • Protection of models, prompts, and training data • Controls against manipulation, poisoning, or misuse • Integration with incident response and crisis management/disaster recovery planning Legal and regulatory alignment As I’ve previously discussed, it seems likely that most AI claims will be evaluated under existing legal frameworks, at least now. • IP ownership and infringement analysis • Compliance with privacy, consumer protection, and sector-specific rules • Processes for tracking regulatory developments Preparing the claims team Claims teams will need to be ready for AI-related claims. • Ability to investigate AI-driven events and decision chains • Access to logs, audit trails, and explainability artifacts • Coordination between claims, legal, underwriting, and tech specialists #AI #cyberinsurance

54

4 Comments

Wilson Sonsini Goodrich & Rosati's 2026 preview outlines key U.S. regulatory developments, including Colorado's AI Act effective June 2026 mandating risk management and impact assessments for high-risk AI, California's Transparency in Frontier AI Act requiring safety frameworks since January 2026, and new CCPA rules for automated decision-making starting 2027. Ignoring these timelines could result in fines up to 7% of global revenue, operational halts, and barred market access. For the safety sector, this escalates demands for documented human oversight in AI-driven hazard assessments and decisions. High-stakes industries like mining must prepare for algorithmic discrimination checks and transparency requirements now. Shema K Solutions sees this as a catalyst for North American governance adoption, mirroring EU pressures. K.I.N.G. framework aligns directly with these mandates by automating audit trails, tiering AI based on risk severity, and enforcing human verification for high-impact uses. It streamlines impact assessments to avoid discrimination, ensuring compliance without disrupting workflows. Watch for Colorado AI Act amendments amid ongoing debates and potential federal interventions. TAKEAWAY: Inventory your AI systems against 2026 regs today—proactive prep averts crisis costs. #AINews #SafetyTech #AISafety #CopiousKing

2

2 Comments

Earlier today I had the pleasure of presenting "AI-Enabled Patent Strategies" to the National Association of Patent Practitioners (NAPP). Big thanks to David Dan Smith (NAPP President) and Josh Snider (Prof. Dev. Committee Chair) for the invitation and for steering a great conversation, to Carol Wilhite (Executive Director) for making the whole thing run so smoothly behind the scenes — and to an audience that kept the Q&A queue full well past the hour. A few threads that seemed to resonate: * The patent bar is not replenishing itself. Average experience is now 21+ years (it was ~16 when I first wrote about this in 2019), new registrations are running at roughly half the rate of the early 2000s, and 62% of those sitting for the exam are now non-attorneys. We are becoming an agent bar — a leaner, more specialized profession facing steady demand. * The economics have quietly inverted. Partner rates have risen ~2% over six years while CPI rose ~23%. Effective rates keep falling, yet client budgets haven't collapsed. That gap creates a rare window: firms that adopt AI can cut prices meaningfully and still expand their margins. The patent right is worth exactly what it was worth three years ago — it just costs far less to produce the documents that secure it. * The provisional is no longer a document production strategy. It's a filing strategy. When a fully enabled application takes hours instead of weeks, "thin provisional" stops being a budget necessity and starts being a disservice to the client. That single shift unlocks approaches most of us have only ever talked about in theory — rolling provisionals, provisional netting, cluster filing — with priority, cost, and optionality no longer in tension. My favorite moment: being asked for actual allowance data. Every AI vendor in this space should welcome that question. Accountability for work product — tracked all the way through the patent office — is where this industry needs to go.  🎥 NAPP made the recording publicly available — it is posted in the comments below. And if you're a patent practitioner thinking about what these shifts mean for your practice, my door is always open. #PatentLaw #IntellectualProperty #LegalTech #AI #PatentProsecution

11

3 Comments

On December 19, NY Gov. Kathy Hochul signed the Responsible AI Safety & Education (RAISE) Act into law, making New York the second state in the nation to codify public safety disclosure and reporting requirements for developers of frontier AI models. My Covington & Burling LLP colleagues Micaela McMurrough, Andrew Longhi, August Gweon and I break down the bill and its impact in our new blog post. #RAISEAct #stateAIlaws #frontiermodels #AIsafety #TFAIA https://lnkd.in/epUUmitW

3

As state and federal AI regulations keep evolving, regulatory change management becomes essential for organizations... 1. Establish Governance and Ownership Ensure regulatory change is clearly owned, resourced and embedded into business operations - Assign regulatory change owner - Create cross functional AI Governance working group 2. Understand business pain points - Connect regulatory requirements to real operational and customer risks. - Map data types, risk levels, and regulated sectors - Collect feedback from team on regulatory friction points 3. Monitor regulatory landscape continuously Detect regulatory change early and translate it into business relevant intelligence. - Regulatory intelligence log - Monthly regulatory change summary - Use technology tools for tracking and monitoring state AI laws, federal agency guidance, enforcement trends, and executive actions. 4. Identify and Assess Regulatory change Decide whether a regulatory development requires action and how urgent it is. • - Regulatory Impact Assessment - Priority rating - Gaps in existing controls 5. Translate Translate legal requirements into business risk, enforcement exposure, and revenue impact. 6. Operationalize Regulatory Changes Embed regulatory requirements into everyday processes - Updated AI policies, AI register - Impact assessment templates - Product compliance checklist 7. Communicate and Train Ensure teams understand and apply regulatory changes correctly - Role based guidance materials - Training sessions or decks 8. Continuous Monitoring and Improvement #regulations #ai #aigovernance #privacy #techpolicy

2