In five months in 2026, the four largest AI companies on earth all moved into legal work. Anthropic launched a legal offering in May, Google in August, OpenAI in September — and on October 15, Microsoft made Copilot for Legal generally available, with 20-plus purpose-built legal skills inside Microsoft 365 Copilot, in Word and its Cowork workspace. The sequence marks the moment legal AI stopped being a startup niche and became a platform battleground. For founders, law firms, and investors, the rules of the category have just changed.
What Microsoft Actually Launched
Copilot for Legal is not a standalone product. It is a set of skills embedded in software firms already buy: contract review against playbooks, drafting from precedents, clause analysis, document comparison, tracked amendments in Word, extraction of obligations and deadlines, and structured memoranda and chronologies. It builds on a Legal Agent previewed in April 2026, with partner plugins from more than two dozen legal vendors, including Bloomberg Law, iManage, NetDocuments, Docusign — and Harvey and Legora, the two leading legal AI startups. The Word plugin is on by default for commercial Copilot subscribers; the Cowork plugin needs admin enablement and is usage-billed.
The point is distribution: Microsoft need only switch on a capability inside Word, where lawyers already spend their days — a fundamentally different sales motion from the one legal AI startups rely on, and why the launch matters more than any single feature. Microsoft reports early-access users kept roughly seven in ten suggestions — company-reported, with no independent testing yet.
The Startups Are Not Standing Still
Yet the two companies most threatened by Big Tech's arrival are also among the most valuable startups in AI. Harvey raised $550 million on September 9, 2026 at a $15.5 billion valuation, taking its total raised past $1.5 billion since 2022. It reports revenue above $400 million a year, more than 3,000 customer organisations and usage by roughly 80% of Am Law 100 firms. Rival Legora answered on October 21 with a $550 million Series E at an $8.8 billion valuation.
Both now build their own models rather than renting them: Harvey shipped Tenet, its first proprietary model, post-trained on Moonshot AI's open-weight Kimi K3. Harvey was valued at $3 billion in February 2025, $5 billion four months later, $11 billion in March 2026, and $15.5 billion by September — roughly 39 times annual recurring revenue, a multiple that assumes Harvey becomes infrastructure for the legal profession, not merely a popular tool.
What Law Firms Are Actually Doing
Adoption data complicates the hype. The International Legal Technology Association's 2026 survey of 500-plus member firms, representing 140,000 lawyers, found 94% using or exploring generative AI — but the tools named were dominated by platforms firms already owned. Copilot led at 76%, with Claude and Thomson Reuters CoCounsel at 44% each and Harvey at 43%. But interest is not rollout: among firms using or exploring Claude, only 7% had fully rolled it out to attorneys.
In October, Bloomberg Law reported Anthropic told law-firm leaders it would not negotiate key privacy and data-security terms without a minimum spending commitment — a reminder that platform giants set terms, and firms take them or leave them. Firms are experimenting broadly and committing narrowly. The winners will be the tools that survive the journey from pilot to firm-wide rollout, which depends less on raw capability than on governance — privilege, confidentiality, audit trails, and who is responsible when the AI gets it wrong. Microsoft says prompts and documents stay within the Microsoft 365 boundary and are not used to train models. That assurance may matter more than a benchmark score.
What It Means for Founders, Firms and Investors
For founders, the message is stark: the base layer of legal AI — drafting, review, summarisation inside the document — is now a platform feature, not a startup. Point solutions that do one generic task slightly better than Copilot will be absorbed or out-distributed. The durable opportunities sit where platforms cannot easily go: deep workflow ownership in a practice area, proprietary data such as a firm's own precedents, integrations with the systems of record, and the judgement layer — knowing what good looks like in a jurisdiction or deal type. Harvey and Legora survive because they are building toward that depth. A startup that rents a frontier model and wraps it in a legal prompt does not.
For law firms, adopt the bundled platform capabilities — but do not mistake them for a legal AI strategy. The hard questions — which matters may touch the tools, how outputs are verified, who owns liability — are governance decisions, to be made before firm-wide rollout, not after.
For investors, the category now demands a sharper underwriting lens. Their valuations assume a winner-takes-most market, yet distribution now sits with Microsoft, Google, OpenAI and Anthropic — also the startups' model suppliers. Diligence: what does the company own that a platform cannot replicate, how sticky is usage after the pilot, and what happens to margins if a platform bundles the same capability free? Legal AI is no longer a bet on whether lawyers will use AI. That question is settled. It is a bet on who captures the value — and that is a much harder bet to win.
Angeled.com connects founders with investors.
Sources: Microsoft Copilot for Legal launch coverage (LawNext, Legal IT Insider, The Professions Cloud), October 15-17, 2026; Tech Funding News, September 10, 2026 (Harvey $550M at $15.5B); TechCrunch / Tech Insider, October 21, 2026 (Legora Series E at $8.8B); ILTA 2026 technology survey via LawSites, September 14, 2026; Bloomberg Law via FourWeekMBA, October 22, 2026.
Read Next
OpenAI Releases Codex Security as Open-Source CLI and SDK
Databricks Unveils Always-On Genie, an AI Agent That Stops Waiting to Be Asked
DeepMind Launches AlphaGenome 2, a DNA Model That Reads Genomes Cell by Cell
ElevenLabs Unveils 13 New Products and Upgrades at Its New York Summit
Google Launches Gemini Decision 1, a Model That Answers Instead of Writing
Runway Open-Sources Praxis-1, Its Robot Model That Learned by Watching Video



