On September 17, 2026, OpenAI announced Astra for Law — GPT‑6 Astra configured as an AI foundation for law firms and legal technology companies. Not “another chatbot for lawyers,” but a stack: a model, a legal search index, instructions for legal analysis and writing, and integrations into the systems firms already use.
For U.S. big law, this is the next round of the legal-tech arms race. For Ukrainian business and the Ukrainian legal profession, it is a different kind of signal: clients, investors, and international counterparties are already adapting to a new speed of legal work. If we miss that, we do not lose to “a model.” We lose to competitors who put AI into process without dissolving accountability.
What OpenAI actually launched
Astra for Law is GPT‑6 Astra tuned for legal practice: search, analysis, drafting. The core addition is a legal search index — more than 230 million URLs covering U.S. case law, statutes, and regulations. In partnership with the Free Law Project / CourtListener, coverage reaches about 99.9% of published U.S. precedential case law.
On the Vals AI Legal Research Bench, the full Astra for Law setup at highest reasoning effort scored 54.0% correct versus 38.7% for GPT‑6 Astra with web search alone — roughly a 40% relative improvement. That is not “magic instead of a lawyer.” It is a measured lift on legal research inside an American corpus.
Access starts via Trusted Access for selected firms in ChatGPT and Codex. The API is coming as GPT‑6 Astra Law / gpt-6-astra-law. For eligible firms: Zero Data Retention on the API; ChatGPT Enterprise excluded from human review by default. Governance work is with Latham & Watkins.
The ecosystem is already wrapping the product: Harvey and Legora on the API; plugins for Relativity, Clio, iManage, Intapp, Thomson Reuters HighQ / CoCounsel and others; ChatGPT for Word is generally available. Firm examples include an agreement analyzer at Sullivan & Cromwell, diligence workflows at Ropes & Gray, and Cooley GO Public. OpenAI is selling not “an answer to a question,” but embedding into firm workflow.
The important caveat for Ukraine
The index is U.S.-centric. A Ukrainian lawyer cannot treat Astra for Law as authority on Ukrainian law, the Criminal Code, Supreme Court practice, or the EU acquis. Pasting a model answer into a pleading “because OpenAI said so” is a professional failure, not innovation.
That does not make the announcement irrelevant. It clarifies several pressures already on our market:
First, the bar for speed and depth of due diligence, contract review, and research on international deals is moving up. A Ukrainian company selling into the EU or U.S., or raising foreign capital, already faces counterparties whose counsel work with tools like this.
Second, confidentiality and governance expectations rise. Zero Data Retention, Trusted Access, exclusion from human review, and big-firm partners become the new client baseline for where matter materials may go.
Third, the legal-tech race accelerates. Ukrainian firms and in-house counsel are not competing with OpenAI. They compete with Ukrainian and foreign teams that convert AI into repeatable process without blurring partner or GC responsibility.
Fourth, the local data gap becomes obvious: Ukrainian legislation, court practice, the Unified State Register of Court Decisions, and EU law in Ukrainian and structured form. Without that, a “global model for law” in Kyiv remains a helper for English-language contracts and general analysis — not a substitute for local expertise.
What this changes for business
I see this in the companies I run and in the ones around them. Legal is still too often treated as a brake: send the contract, wait a week. In a world where firms like Sullivan & Cromwell and Cooley embed models into agreement analysis and public products, a week on a standard NDA looks like a competitive weakness, not prudence.
This is not a call to fire lawyers. It is a call to change the work. A business that wants to close deals faster, check counterparties, prepare a data room, and answer diligence must ask: which legal tasks are repetitive, where can we speed a first draft layer, and who owns the final “yes.”
For a founder or CEO the practical effect is simple. If your GC or outside firm has no AI policy, you already take two risks: either slow cycle time, or leakage of confidential material through personal ChatGPT accounts without enterprise controls.
What this changes for the legal profession
Junior lawyers who only know how to “Google and rewrite a template” will feel pressure first. Senior partners who can frame questions, price risk, and own the call become more valuable — if they teach the team to use tools without the illusion that the model “knows Ukrainian law.”
Firms that wait for “a Ukrainian CourtListener with 99.9% coverage” will lose to firms that already build internal playbooks: which documents may go through models, which only in isolated environments, who does human review, how usage is logged, how associates are trained to verify citations and norms.
In-house is a separate story. In a typical Ukrainian company the legal team is small. That is exactly where AI has leverage: first pass on supply contracts, NDAs, employment templates, redline comparison, preparing questions for outside counsel — if the GC sets the frame, instead of “dropping ChatGPT into a messenger.”
What to change now
Not a three-year strategy. The next moves.
1. Adopt an AI policy for legal work
One two-to-three-page document: what data is allowed and forbidden; which tools are permitted (enterprise / ZDR / local); who approves external vendors; a ban on putting client matter into personal accounts. Without that, every “pilot” is a lottery.
2. Separate “draft” from “decision”
A model can prepare a first overview, a comparison table of redlines, a risk list in a contract, a draft letter. The decision belongs to a lawyer with a signature. Encode that in process as strictly as you encode who approves a payment.
3. Start with English-language and repetitive flows
International NDAs, English MSAs, diligence questionnaires, redline comparison — where corpus and language are closer to what frontier models are trained on. Do not start with a unique litigation strategy in a Ukrainian commercial dispute.
4. Demand transparency from outside counsel
Ask whether they use AI; on what infrastructure; whether ZDR / enterprise applies; who reviews. If a firm stays silent or claims “everything is manual” while invoices grow — that is already a maturity signal, not a quality badge.
5. Invest in local data and knowledge
Internal templates, firm position playbooks, prior deal libraries, structured opinions — that is your real “legal index.” Without it, a global model runs idle. With it, it accelerates your practice specifically.
6. Train people, not only licenses
Teach associates to verify citations, norms, and facts. Teach partners to brief a model. Teach GCs to measure cycle time on a standard contract before and after. A tool without skill is decoration in a board deck.
7. Watch EU/UA regulation and client NDAs
The AI Act, GDPR, banking and medical secrecy, client NDA terms — all already constrain where text may go. Astra for Law is interesting as a controls baseline; your policy should be stricter than a vendor slide.
What the state and ecosystem should do
Briefly: without high-quality open and licensed corpora of Ukrainian law, local legal AI will remain an import of Anglo-American logic. We need machine-readable acts, stable access to court practice, incentives for legal tech, and clear bar rules on AI use that do not dissolve attorney–client privilege.
This is not “IT hype.” It is trust infrastructure for the legal system in a digital age — alongside court reform and enforcement of judgments.
Bottom line
Astra for Law is an American product on an American legal corpus. The signal is global: legal work is becoming a measurable engineering discipline — with indexes, benchmarks, plugins, and governance. Ukrainian business and Ukrainian lawyers can ignore the OpenAI brand. They cannot ignore the shift in speed and standards this announcement locks in.
Do not wait for a “Ukrainian Astra.” Write the policy. Separate draft from decision. Accelerate the repetitive. Keep human accountability. And in parallel demand local datasets — or we will keep importing someone else’s infrastructure instead of building our own.
OpenAI’s original announcement is here: openai.com/index/astra-for-law.