An NDA arrives. It's labeled "standard mutual NDA." The counterparty says it's the same agreement they use with everyone. Counsel has 20 minutes before the call. The document gets a quick scan, not a full review.
That's how non-competes end up in signature lines. Not from carelessness. From volume and the assumption that "standard" means reviewed.
In a benchmark test of 20 NDAs, extraction surfaced 6 minutes of structured output per document. The provisions that appeared most often in unexpected locations: embedded non-competes, IP assignment clauses, and 5-year survival terms. None of these were flagged in the cover email as non-standard.
Upload your NDAs. See every provision on one screen. Free workspace, no card required.
Extract my NDAs freeWhy NDAs Get Under-Reviewed
Three reasons. First, volume: a mid-market legal team signs 50 to 100 NDAs per year (typical range for active deal teams). At a 30-minute review per document (industry norm), that's 50 hours annually on routine agreements. In practice, review time gets compressed to 5 to 10 minutes.
Second, framing: "standard NDA" primes reviewers to look for deviations from a mental template rather than reading the document itself. The assumption is that it's already been reviewed by someone else.
Third, placement: risky provisions rarely appear in the main body. They appear in definition sections, survival clauses, and exhibit schedules where attention drops off.
Five Provisions That Hide in "Standard" NDAs
Embedded non-competes. Buried in the definition of "Confidential Information" or in a separate "Restrictions" section. Often written as: "Each party agrees not to solicit or hire employees of the other party or engage in any business activity that directly competes with..." This is not a confidentiality provision. It's a non-compete.
Broad IP assignment. "Any ideas, inventions, or improvements conceived during the term of this agreement relating to the Disclosing Party's business shall be assigned to the Disclosing Party." This can assign IP developed independently if it touches the disclosed subject matter.
5-year or indefinite survival. Standard NDAs have 2- to 3-year survival clauses. Some use "perpetual" or omit the term entirely. Perpetual survival on trade secrets may be reasonable. Perpetual survival on all confidential information is not.
Unilateral injunctive relief. "The Disclosing Party shall be entitled to seek injunctive or other equitable relief without the requirement of posting bond." This waives the receiving party's right to contest injunctive relief. Often buried in a "Remedies" section near the end of the document.
Residuals clauses. "Notwithstanding the foregoing, the Receiving Party may use Residual Information in the ordinary course of business." "Residual Information" is often defined as information retained in unaided memory after authorized use. This can gut the NDA's practical enforceability.
The Extraction Approach: Every Clause, Every Page
Extraction builds a grid of every provision across every NDA. Columns include: party type (mutual or unilateral), term, survival period, permitted disclosures, IP clauses, non-solicitation, governing law. Every cell cites the page and section.
The value isn't in the AI reading the document. It's in seeing all 20 NDAs on one screen, with every material provision visible without opening a single file. Deviations from standard terms are visible immediately.
A Benchmark: 20 NDAs, 6 Minutes
In our benchmark test with 20 sample NDAs (ranging from 3 to 18 pages), extraction surfaced the following across the portfolio: 4 NDAs with embedded non-compete language, 7 with survival periods exceeding 3 years, 3 with residuals clauses, and 2 with unilateral IP assignment provisions. Total extraction time: 6 minutes. Total manual review at 20 minutes per document: 6.7 hours.
Of the 4 non-compete provisions found, none were flagged in the document titles or cover emails. All were in sections titled "Confidential Information" or "Restrictions."
What Extraction Misses in NDAs
Extraction does not catch provisions in tracked-changes markup that hasn't been accepted. It does not read side letters or verbal amendments. It will not surface carve-outs that require cross-referencing two separate agreements.
Extraction gives you a complete map of the signed document. What happens outside the four corners of that document requires a different review.
Further reading: NDA data extractor · Eudoxic for corporate teams · Hidden liability in contracts