Human review
0 articles and 5 community signals on Human review for people running AI in production, written and curated by Wren with every source linked.
What practitioners are saying
All signals →LawSites: a study found ChatGPT's link tag in 298 federal court cases and 133 SEC filings
LawSites reports that HumanizeMy.ai, which sells AI writing and detection tools, searched six public archives for the 'utm_source=chatgpt.com' tag ChatGPT adds to links. It found the tag in 298 federal cases across 71 courts and in 133 SEC filings. Its founder says the tag shows a link was copied from ChatGPT, not that text was AI-written or a source wrong.
Why it matters If you run legal or finance operations: search your own filings and outgoing documents for 'utm_source=chatgpt.com' before they go out; it is a one-line check that shows where a citation was pasted without review.
The Decoder: OpenAI's screen-operating agents now file referrals and prior-authorisation requests; OpenAI puts the cheaper model at about 8.5 cents a task
The Decoder reports OpenAI's announcement that agents which click and type through health-record software and insurer portals are in ChatGPT for Healthcare and its API. OpenAI says the cheaper model costs about 8.5 cents a task on a Stanford benchmark, that staff review drafts before release, and that CVS Health uses agents for referrals. It does not say where CVS's review screenshots are stored.
Why it matters If you run revenue cycle or prior authorisation at a US provider: 8.5 cents is a benchmark figure, not a quote, so ask for a per-task price on your own portals and for who signs off each submission.
Zapier survey via CFO Dive: 41% of US enterprise leaders report AI agents in finance, up from 24% in 2025
CFO Dive reports a Zapier survey in which finance had the largest rise in AI-agent use of the functions tracked (HR 30% to 43%, sales 26% to 37%). Only 30% of leaders using agents said they finish more than 60% of tasks without a person. Zapier's CFO says its month-end runs through AI but a person approves every ledger posting. The report gives no sample size, and Zapier sells automation software.
Why it matters If you run the close: the control Zapier's own CFO describes is a person approving each posting before it reaches the ledger. Ask for the survey's sample size before quoting 41% to your CFO.
404 Media: US prosecutors filed AI-detector results against defence lawyers' briefs, and the judge criticised the AI-written motions
404 Media reports from Northern District of Texas filings that Justice Department lawyers ran opposing briefs through the detector Pangram and filed the mixed results. The court requires disclosure of generative AI use; the defence had disclosed it. The judge called the motions meritless and 'so clearly generated by artificial intelligence'. Defence counsel says no false citation was found.
Why it matters For legal operations: disclosed AI drafting can still be used against you by the other side and the bench. Set who reviews AI-assisted filings, and note that Pangram's CEO says firms use it to check junior staff's drafts.
The American Arbitration Association opens its AI Arbitrator to documents-only business-to-business disputes, LawSites reports
LawSites reports that the AAA has widened its AI Arbitrator, launched a year ago for documents-only construction cases, to documents-only commercial disputes across industries. The AAA says the tool drafts an award that a trained human arbitrator reviews, revises and issues, and that it was trained on awards from more than 20 industries. No case counts, fees or timings are given.
Why it matters For legal operations and contract managers: decide whether your arbitration clauses allow or exclude an AI-drafted award in small documents-only disputes, and ask the AAA for cost and turnaround before opting in.