Lead story
DeepJudge Publishes Agent Handoff Protocol as Harvey and Thomson Reuters Join Early Adoption
DeepJudge released a draft open protocol for transferring a user, task objective, selected conversation context, and approved resources between independently operated AI applications.
- What happened
- DeepJudge released a draft open protocol for transferring a user, task objective, selected conversation context, and approved resources between independently operated AI applications.
- Why it matters
- Moving matter context and files between legal AI products creates a new information-governance boundary. Legal and discovery teams evaluating handoffs should require user confirmation, data minimization, authorization, retention controls, durable audit records, validated destinations, and clear rules for preserving transferred materials and resulting work product.
- Watch
- Published August 13. AHP is a draft v1 specification, Harvey's implementation is planned for beta, and Thomson Reuters has not announced its implementation timing; interoperability and security performance have not been independently evaluated.
Sourcesartificiallawyer.comgithub.com
Briefing item
DISCO Makes Advanced Research Generally Available for Multi-Step Evidence Analysis
DISCO announced general availability of Advanced Research, an agentic AI application intended to perform autonomous, multi-step analysis of litigation evidence while showing its reasoning and logic.
- What happened
- DISCO announced general availability of Advanced Research, an agentic AI application intended to perform autonomous, multi-step analysis of litigation evidence while showing its reasoning and logic.
- Why it matters
- Agentic evidence research can influence investigative direction, witness preparation, and case theories. Teams should validate answers against cited records, preserve queries and outputs, test permission and privilege boundaries, document human review, and determine whether research sessions and intermediate results fall within retention, legal-hold, or production obligations.
- Watch
- Reported August 14 in a multi-item industry roundup. The available report provides a concise vendor announcement rather than independent accuracy testing, implementation details, or comparative performance data.
Sourcesartificiallawyer.com
Briefing item
Anthropic Details Machine-Readable Marks for Claude-Generated Content
Anthropic said supported Claude outputs will use imperceptible text watermarks and signed provenance metadata for supported files as it implements EU AI Act transparency commitments.
- What happened
- Anthropic said supported Claude outputs will use imperceptible text watermarks and signed provenance metadata for supported files as it implements EU AI Act transparency commitments.
- Why it matters
- Machine-readable provenance may provide an additional signal when authenticating or investigating AI-assisted content, but it should be preserved with native files and metadata rather than treated as conclusive proof. Discovery protocols may need to address collection methods that retain provenance labels and testing of detection tools used to support evidentiary claims.
- Watch
- Published August 13 and based on Anthropic guidance updated this week. Anthropic says technical detection documentation is forthcoming; marks can have coverage and persistence limits, and the presence or absence of a supported mark does not by itself establish authorship or authenticity.
Sourcesartificiallawyer.comsupport.claude.com
Briefing item
Digital Forensics Guidance Adds AI Prompts and Outputs to Preservation Planning
New digital-forensics guidance identifies prompts, generated content, and AI-assisted workflows as potential evidence sources and describes AI-assisted methods for prioritizing complex investigation data.
- What happened
- New digital-forensics guidance identifies prompts, generated content, and AI-assisted workflows as potential evidence sources and describes AI-assisted methods for prioritizing complex investigation data.
- Why it matters
- Prompt histories and generated outputs may be missed when legal holds and collection plans focus only on conventional email and documents. Practitioners should inventory enterprise and individual AI services, test available exports, preserve metadata and account context, and document how automated prioritization was validated before relying on it in an investigation.
- Watch
- Highlighted by eDiscovery Today on August 13. The underlying article is practitioner guidance from Veracity Forensics, an eDiscovery Today sponsor; examples and workflow benefits are not independent empirical findings, and AI-generated signals can produce false positives or miss context.
SourceseDiscovery Todayveracityforensics.com
Briefing item
Spellbook Adds Integrated AI and Manual Editing Across Contract Sets
Spellbook launched an AI document editor that combines automated redlining with manual edits, comments, formatting, citation navigation, and review of changes inside its Associate environment.
- What happened
- Spellbook launched an AI document editor that combines automated redlining with manual edits, comments, formatting, citation navigation, and review of changes inside its Associate environment.
- Why it matters
- Integrated editing environments can become repositories for document versions, comments, prompts, citations, and approval decisions that may later be relevant to disputes or investigations. Organizations should confirm version-history retention, native export fidelity, authorship metadata, access logs, legal-hold support, and whether rejected AI changes remain recoverable.
- Watch
- Reported August 14, with the product page last published August 13. Feature descriptions and workflow examples are vendor claims; the sources do not provide independent accuracy, security, or export-fidelity testing.
Sourcesartificiallawyer.comspellbook.com
Educational only. Not legal advice. Review sources and matter-specific obligations before relying on any workflow decision.