Sullivan & Cromwell's Agreement Analyzer: what a bespoke, in-house redline tool actually means for the rest of us

What it is

Together with OpenAI, Sullivan & Cromwell announced the collaborative development of Agreement Analyzer, a custom application that integrates S&C's workflows with OpenAI's ChatGPT Enterprise. The tool was included in the launch of OpenAI's Astra for Law. Mechanically, the S&C collaboration produced the Agreement Analyzer, a tool that runs contracts through multiple automated review cycles to flag issues, check language against the firm's own negotiating playbooks, and identify potential litigation risks. It started small: Sullivan & Cromwell partner John Savva developed Agreement Analyzer as a single-user proof of concept built with Codex, then was hardened with OpenAI's engineers.

Important: this is an internal S&C build, not a product you can buy.

The next-work-session test

Imagine your Monday: a 60-page master services agreement lands in your inbox, counterparty paper, due Wednesday. You need to compare its indemnity, limitation-of-liability, IP, and termination language against your group's standard positions, flag deviations, and draft redlines.

Agreement Analyzer, as described, is designed for exactly that: ingest the draft, run it against the firm's playbook and precedents, surface issues, propose markup. If you're an S&C associate, your next session changes — the first-pass triage is machine-done and you spend your time on the judgment calls. If you're outside S&C, your next session is unchanged. You cannot license this tool. The relevant takeaway is the pattern, not the product.

Pricing

Pricing: unverified — and in a specific way. Agreement Analyzer itself is not for sale. The Agreement Analyzer was developed for internal use at S&C. It runs on ChatGPT Enterprise, which OpenAI prices on a custom basis per organization (no public per-seat rate). For eligible firms, the offering includes Zero Data Retention (ZDR) on our API, and usage of ChatGPT Enterprise is excluded from human review by default under OpenAI's Astra for Law Trusted Access Program — but access terms and cost for that program are not published.

If you want the capability without the S&C badge, the real cost is: ChatGPT Enterprise seats (quoted) + your own engineering build + playbook curation. OpenAI's own framing is that OpenAI has shared live examples of how firms can turn their own precedents, methods, and judgment into AI tools and workflows built to their standards — i.e., you build it.

What we'd actually use it for

The honest, narrower version for a legal team that isn't S&C: use ChatGPT Enterprise (or a comparable enterprise-grade LLM with ZDR) to run a draft against a single clause playbook you've written down — say, your 12 standard indemnity positions. First-pass deviation spotting. A memo to yourself before you open Word.

Not: autonomous redlining a counterparty draft and sending it out. The S&C framing is deliberately modest here too — a custom application that integrates S&C's workflows with OpenAI's ChatGPT Enterprise is a workflow tool, not a replacement associate.

Limits

• You can't buy it. Repeat because it matters. Every write-up calling this a "product launch" is wrong.
• It's only as good as the playbook you feed it. S&C's edge is decades of curated precedent and partner-authored positions. A generic team pointing GPT at a folder of old NDAs will not get S&C output.
• Privilege and confidentiality are a configuration problem, not a feature. ZDR on the API is available under the Trusted Access Program, but your conflicts, client-matter segregation, and outside-counsel guidelines are still your problem.
• Issue-spotting ≠ redlining. The public descriptions emphasize flagging issues and checking against playbooks. Clean, defensible track-changes markup in Word is a separate workflow problem.
• GPT-6 Astra is the underlying model per the launch materials, and model behavior on long contracts (hallucinated cross-references, missed defined terms) remains a known failure mode across every legal LLM tool. Human review is non-negotiable.

Try it if

• You're at S&C. (Obviously.)
• You're a GC or firm with budget for ChatGPT Enterprise + an engineer, and you have a written playbook worth encoding.
• You want a reference architecture for arguing internally that bespoke beats off-the-shelf contract review tools for your specific practice.

Skip it if

• You were hoping to license Agreement Analyzer. You can't.
• Your "playbook" is tribal knowledge in three partners' heads. Fix that first; the tool question is downstream.
• You need redline output in Word with clean track changes today — look at purpose-built contract review vendors instead.
• You can't get Zero Data Retention terms in writing from your LLM provider. Don't put client paper through anything less.

The real story in OpenAI's write-up isn't the tool. It's that a top firm decided the shortest path to useful AI was building their own on top of ChatGPT Enterprise rather than buying a legal-AI SaaS. That's the decision worth copying — not the software.

---

This content is for informational purposes only and is not legal advice. Confirm confidentiality, privilege, and jurisdictional rules before using any AI tool with client matters.