Matter isolation and privilege protection on
Law firms work under the strongest confidentiality obligations in professional services, yet 44% still have no formal AI governance policy while 79% of legal professionals already use AI. Igris isolates each client matter at the gateway, redacts identifying details before any prompt reaches an LLM, and produces the supervision record ABA Rule 1.6 and 5.3 require.
AI Contract Review
A firm that deploys LLMs to review and summarise contracts is processing client names, counterparty identities, deal structures, transaction values, and confidential commercial terms — information protected not just by data protection law but by ABA Model Rule 1.6, which imposes a professional ethics obligation distinct from and additional to any regulatory requirement. If client information from Matter A is accessible in the context of a prompt sent while working on Matter B, the firm has breached attorney-client privilege and created a Rule 1.6 ethics exposure that a state bar investigation can open on the basis of a single client complaint. The consequences of an ethics finding are not a regulatory fine. They are a disciplinary proceeding, a potential malpractice claim, and a client relationship that does not survive the disclosure.
Igris creates an isolated connection for each client matter, so no context, no prompt history, and no data from one client's work can appear in another's — matter isolation enforced at the gateway layer. Guard PII redaction strips party names, financial figures, and transaction identifiers from prompts before they leave the firm's network. Encrypted body logging covers the firm's compliance record — every LLM interaction involving client matter work, retained according to the firm's document retention policy. Igris Lens tracks cost per matter connection, giving the billing team clean per-matter AI spend data that feeds directly into client billing.
Every client matter is isolated at the governance layer. ABA Rule 1.6 confidentiality obligations are met by a technical control that can be described to a client or demonstrated to a bar ethics committee. Per-matter cost tracking makes AI spend a recoverable billing input rather than a firm overhead cost that erodes margin invisibly.
Legal Research Assistant
Associates using AI tools to research case law and draft legal memoranda work within the context of active client matters — which means their prompts naturally contain case numbers, client identities, opposing counsel identities, factual summaries, and in some workflows, full deposition transcripts. ABA Model Rule 1.6 makes the protection of that information a professional ethics obligation. ABA Model Rule 5.3 makes supervising partners responsible for ensuring that AI systems are used in ways consistent with the firm's professional obligations — which means the partner is accountable for what the associate's prompt contained. When an associate pastes a full deposition into an LLM prompt, the supervising partner may be unaware that a Rule 5.3 compliance gap has occurred until a client asks what happened to their information.
Guard content guard policies use custom detection patterns built for legal contexts: case number formats, court filing identifiers, client matter codes, and the naming conventions the firm uses internally for parties and counsel. When an associate's prompt contains any of these, Guard redacts them before the prompt reaches the LLM provider. Token limits prevent associates from pasting complete deposition transcripts as context. The Igris audit trail records every research query, every redaction event, and every model used across all associate interactions — giving managing partners the supervision record that ABA Rule 5.3 requires.
Associates get the research speed of AI assistance without the privilege risk of uncontrolled prompt content. Managing partners have a supervision record that satisfies ABA Rule 5.3 obligations automatically. When a client asks how their case information was handled during AI-assisted research, the firm has a documented answer.
AI Due Diligence Assistant
M&A due diligence is the most information-dense and commercially sensitive work in corporate law. The data room contains target company financials, employment records, IP assignments, regulatory filings, pending litigation details, and proprietary business information that is confidential under the transaction's NDA, protected under ABA Rule 1.6, and covered by GDPR Article 9 for any European employee data in the records. Multiple parties operating in the same transaction create a data commingling risk that is structural to the process. A privilege breach in an M&A matter is not measured in compliance fines — it is measured in deal exposure.
Igris creates a dedicated connection for each transaction, isolating the due diligence AI workload from every other matter at the gateway layer. Guard applies PII redaction to employee records and personal data within due diligence documents, addressing GDPR Article 9 obligations for EU employee data and DPDP Act 2023 obligations for Indian employee records. Custom content patterns catch deal-specific codenames, transaction identifiers, and proprietary financial terms before they reach the LLM provider. Sentinel governs every tool call the due diligence agent makes, enforcing an allowlist that restricts which data room sections the agent can access in a given session.
Due diligence AI runs within a governance perimeter that matches the sensitivity of the transaction. No deal data crosses matter boundaries. The client can ask how their transaction data was protected and receive a documented log rather than an assurance.
AI-Assisted eDiscovery and Privilege Review
Law firms using AI to classify discovery documents work at a volume that makes manual prompt oversight impossible. The privilege review scenario creates a specific and severe consequence: if an attorney-client privileged document reaches an external LLM provider as part of a classification prompt, the firm may have inadvertently waived the privilege. Courts have found inadvertent waiver even when firms took reasonable precautions — the standard is not intent, it is what controls were in place. ABA Rule 1.6 obligations apply to every document in the review set. ABA Rule 5.3 supervision requirements apply to the AI tool conducting the review. GDPR Article 5 data minimization obligations apply to any personal data appearing in the documents.
Guard applies a privilege-aware content policy that detects the markers of attorney-client communications — attorney name patterns, matter reference codes, and privileged communication headers — blocking those documents from reaching the LLM before the call is made. The allowed model list restricts the review pipeline to LLMs approved for privileged document handling. Token limits cap the document size processed in a single prompt. The full audit trail records every document processed, every privilege flag detected, every policy action taken, and every model used.
Privileged documents are caught and blocked before they reach any external provider — not reviewed after the call, blocked at the point of transmission. The supervising attorney has a governance log of every document the AI touched. The firm can respond to a privilege waiver challenge with documented technical evidence of the controls in place.
Give every matter its own governance perimeter
See how firms isolate client matters, block privileged material before it reaches a provider, and track recoverable AI spend per matter.