California bars attorneys from blaming generative AI as NIST profile guides vendor risk work

California closes the "AI made me do it" defence for legal practitioners

Senate Bill 574, signed by Governor Gavin Newsom on September 30, 2026, ends the ability of attorneys and arbitrators to attribute faulty work to generative AI. The law frames a kind approach to attorneys' and arbitrators' use of generative AI and insists on human judgment over machine output. For practitioners, the practical change is straightforward: every filing, brief or arbitration submission that touches a generative system now carries an explicit duty of human oversight that the statute will not let a lawyer delegate to a model. The rule reshapes how law firms document prompts, drafts and reviews.

NIST AI 600-1 sets the baseline for generative risk work

A separate explainer restates the role of NIST AI 600-1, the Generative AI Profile published in July 2024 as a companion to the NIST AI Risk Management Framework. The profile names 12 risks unique to or made worse by generative AI and gives vendors a checklist for governance, intellectual property handling and provenance. For readers who build or buy generative tooling, the profile is the document to map internal controls against, since California's new statute for lawyers and the federal profile converge on the same theme: a generative system cannot be left to act unsupervised, and its outputs cannot be presented without human judgment.

Industry framing reinforces strategic, not reflexive, adoption

A photographer-industry discussion of generative AI captures the wider mood, calling the technology a source of transformative potential that demands strategic prioritization, risk management and a holistic organizational approach. The point for a practitioner audience is that adoption is no longer a creative question alone; it is a governance question. Teams that have not yet mapped their generative usage against the NIST profile, or that lack a written human-oversight policy for legal and customer-facing work, now have two independent reasons to start: a state statute aimed at lawyers and a federal profile that vendors are being measured against.

What practitioners should do before the next deadline

The closest dated action item in the evidence is the September 30, 2026 signing of SB 574; no further compliance deadline is named in the available material. The responsible next step is a documented policy that ties each generative workflow to a named human reviewer, prompt logs that can be produced on demand, and a gap analysis against NIST AI 600-1's 12 named risks. For teams that need to test or anonymise data before any of this work, the Dummy File Generator and Random IP Address Generator support the kind of mock-data hygiene that provenance rules increasingly expect.

Identifiers used inside those test rigs can be generated through the MAC Address Generator and ULID Generator so synthetic records stay sortable. Teams that have to label or watermark their own outputs can sketch the workflow with the Bulk QR Code Generator, and any list-cleanup or reordering step in an evidence trail can be sanity-checked using the guide on how to randomise a list online without losing data.