AI Dive · No. 046

    AI, Security and the Future of Law

    Law and Governance

    Dive 046 in the Law and Governance track, examining Legal AI, Confidentiality, Data Retention, Ethics.

    • Legal AI
    • Confidentiality
    • Data Retention
    • Ethics
    • Securities Law

    § 01Overview

    BackTier Analysis — October 2026

    Story at a Glance

    • AI in legal practice has moved from "whether" to "how safely." Florida Bar Ethics Opinion 24-1 permits generative AI with conditions on confidentiality, supervision and billing.
    • Retention, not training, is the live risk. Preservation orders in the NYT v. OpenAI litigation reached consumer chat logs but not Zero Data Retention customers.
    • The safest architecture limits what the AI can see: one AI identity, one project per matter, redacted files only, no connection to client originals.

    § 02What Changed

    Through 2026, vendor policy and court orders started to diverge. Anthropic imposed 30-day mandatory retention on its frontier models, overriding prior ZDR terms. Google Vault brought Gemini conversations under litigation holds. OpenAI previewed a privacy-preserving safety system that has not yet been independently audited. Each is a vendor statement first; only the contract a firm signs makes it binding.

    § 03Why It Matters

    A deleted chat is not necessarily gone. Data that was never retained is the only data a court cannot order preserved. For a securities practice handling MNPI and Regulation FD questions, that turns AI security into an architecture problem, not a policy memo.

    § 04Why It Matters for AI Visibility

    The same firms that must restrict what AI sees internally also need AI systems to see them clearly externally. Public, verified educational content — research summaries, regulatory explainers, transcripts — is exactly the material that is safe to run through AI and that answer engines cite. Confidentiality inside, clarity outside.

    Practical Rules

    • Ask of every vendor claim: what does the agreement say, and who verified it?
    • Redact first, then use AI. If it cannot be redacted, it stays out.
    • Document AI use at the matter level before a client or court requires it.

    Read the full guide: AI Security for Law Firms.

    § 05Sources

    • The Florida Bar, Ethics Opinion 24-1 (January 19, 2024). https://www.floridabar.org/etopinions/opinion-24-1/
    • BackTier client briefing, "AI, Security & the Future of Law" (October 2026). https://gamma.app/docs/AI-Security-the-Future-of-Law-p0i9td1hqcs2pb8

    Full dive

    The complete dive — figures, framing, and conclusions — is published as a deck. It opens in a new tab; this page stays the canonical on-site record.

    Open the full deck ↗

    Continue the series(use ← / → keys)

    Related dives

    All divesaidive.online ↗