The record for this issue
Three claims and their fine print.
74 claims in the draft, each tested by a second agent against the page it cites. 66 held; 8 did not and were fixed or cut before this went out.
-
01
titleconfirmed
The fine print did the work.
The page says “Headline framing; no standalone factual claim. Consistent with corrections below”
-
02
dekconfirmed
A Utah AI writes acne prescriptions
The page says “technology that "autonomously issues orders for first-time topical acne prescriptions" (Sec. 19(F))”
-
03
dekwrong
but a doctor signs each one first
The page says “Stage 2: "AI submits directly; overseeing physician retrospectively reviews all cases at least weekly."”
-
04
dekconfirmed
An avatar fooled people who weren't told to look
The page says “participants "have a short video call with the model without being told what it is"; "26 believed their partner was a real person"”
-
05
dekconfirmed
Mistral's lead holds, with a qualifier
The page says “"significantly outperforming any open-weight model developed in the US or Europe"”
-
06
summaryconfirmed
Three AI claims from the past week (Utah prescribing pilot, video avatar study, Mistral open-weight model)
The page says “Claims dated Oct 1 (Tavus), Oct 5 (Interesting Engineering), Oct 6 (Tech.eu), all within 8 days of 2026-10-07”
-
07
The storyconfirmed
Documents behind the claims: a Utah pilot agreement, a startup's own study, a model launch post
The page says “"Regulatory Mitigation Agreement"; Tavus: "For every new model, we run a study"; Mistral launch post Oct 6”
-
08
The storyconfirmed
Lost on the way: a qualifier about where the rival models were built
The page says “"developed in the US or Europe"; "leads open-weight models developed outside China by a wide margin"”
-
09
The correctionconfirmed
Utah launched world's first system prescribing on its own, without human doctors (paraphrase of IE headline)
The page says “"Utah launches world's first autonomous prescription system without human doctors"”
-
10
The correctionconfirmed
Interesting Engineering headline, Oct. 5
The page says “"Oct 05, 2026 06:18 PM EST"”
-
11
The correctionconfirmed
Overstated
The page says “Physician "reviews and approves every prescription before submission" in Stage 1; verdict follows from agreement”
-
12
The correctionconfirmed
Utah's Oct. 5 announcement approves Nolla Health for a pilot
The page says “Dated October 5, 2026; Nolla "was approved for a targeted pilot to assess AI-assisted visual evaluations and treatment pathways"”
-
13
The correctionconfirmed
State agreement runs 12 months
The page says “"effective for twelve (12) months beginning on the Commencement Date" (Sec. 2(A))”
-
14
The correctionconfirmed
Data meant to inform future rules
The page says “Data "will be shared with the Utah Legislature and the public" "to inform future, permanent, evidence-based healthcare regulations."”
-
15
The correctionconfirmed
Covers first-time and renewal prescriptions of topical acne treatments for adults
The page says “"first-time topical acne prescriptions and renewals to eligible Utah residents age 18 years and older" (Sec. 19(A))”
-
16
The correctionconfirmed
Mild to moderate acne
The page says “Limited to mild to moderate acne (Schedule B)”
-
17
The correctionwrong
from a list of five drugs
The page says “Section 4.3 table lists 8 formulations built on five active ingredients (tretinoin, adapalene, clindamycin, benzoyl peroxide, azelaic acid); the document never says five medications”
-
18
The correctionconfirmed
No oral medication
The page says “"No oral medications, systemic hormonal therapies, or isotretinoin will be prescribed as part of this pilot."”
-
19
The correctionconfirmed
A physician approves every prescription for at least the first 100 patients
The page says “Stage 1: minimum 4 weeks and 100 patients; "overseeing physician reviews and approves every prescription before submission"”
-
20
The correctionconfirmed
Each later stage, with lighter review, needs the state's written approval
The page says “"Participant shall not commence any succeeding phase without the prior written approval of the Office." (Sec. 20(C)(1))”
-
21
The correctionconfirmed
Doctor steps back only if AI agrees with physicians at least 95% of the time
The page says “Gate criteria: "≥95% concordance"”
-
22
The correctionconfirmed
and misses no hard stop
The page says “"0 missed hard stops after QA"”
-
23
The correctionconfirmed
Hard stop examples: a patient under 18, a pregnancy
The page says “"Patient age under 18"; "Self-reported pregnancy, aiming to become pregnant, or breastfeeding"”
-
24
The correctionconfirmed
Source names map to sources[]
The page says “All four names present in sources[]; primary source listed first”
-
25
The correctionconfirmed
How the doctor steps back / Stages in Nolla Health's agreement with Utah
The page says “Agreement sets Stages 1-3 with reducing physician review”
-
26
The correction, step 1confirmed
A physician approves every prescription, for at least 100 patients
The page says “Stage 1: "Minimum ... 100 patients"; physician "approves every prescription before submission"”
-
27
The correction, step 2confirmed
The AI sends prescriptions, reviewed weekly afterward
The page says “Stage 2: "AI submits directly; overseeing physician retrospectively reviews all cases at least weekly."”
- 28 unsupported
-
29
The correction, step 3confirmed
A monthly review of at least 10%
The page says “Stage 3: monthly sampling reviews "covering a minimum of 10% of enrollment volume"”
- 30 unsupported
-
31
The correctionconfirmed
Each stage has a minimum length
The page says “Stage 1 min 4 weeks/100 patients; Stage 2 min 8 weeks/500; Stage 3 min 750 patients (no minimum weeks listed for Stage 3; holds only if 'length' includes patient counts)”
-
32
The correctionconfirmed
next one can't start without the state's written approval
The page says “"shall not commence any succeeding phase without the prior written approval of the Office"”
-
33
The correctionconfirmed
Tavus says Griffin is the first to pass the real-time video Turing test
The page says “"Griffin is the first model to pass the real-time, video Turing test."”
-
34
The correctionconfirmed
Tavus research post, Oct. 1
The page says “"October 1st, 2026"”
-
35
The correctionconfirmed
Tavus tested Griffin-Lite
The page says “Griffin-Lite described as "a research preview"; n=54”
- 36 unsupported
-
37
The correctionconfirmed
54 people recruited through a research platform
The page says “"recruited through an independent research platform"; n=54”
-
38
The correctionconfirmed
Told they'd be matched with another participant for a one-minute video call
The page says “"would be matched with another participant for a one-minute video call"”
-
39
The correctionconfirmed
26 of them (48%) said their partner was a real person
The page says “"26 believed their partner was a real person" (48%)”
-
40
The correctionconfirmed
Tavus counts that as passing and sets no threshold
The page says “Page claims a pass and never defines a cutoff; earlier systems described as "a max 2% pass rate"”
-
41
The correctionconfirmed
On its older Phoenix-4.5
The page says “"Previous systems, including Tavus's own industry leading conversational video interface"”
-
42
The correctionconfirmed
2.4% of 41 participants said the same
The page says “Phoenix-4.5: "2.4%" (n=41)”
-
43
The correctionconfirmed
Tavus ran the study itself
The page says “"For every new model, we run a study with participants recruited through an independent research platform"”
-
44
The correctionconfirmed
Griffin-Lite is open only to select trusted testers
The page says “available to "select trusted testers as a research preview"; "will not be available for use for customers at this time"”
-
45
The correctionconfirmed
Nobody was asked to look for an AI
The page says “"without being told what it is"”
-
46
The correctionconfirmed
Said their partner was a real person / Tavus's own one-minute call studies
The page says “"For every new model, we run a study"; one-minute call”
- 47 confirmed
- 48 confirmed
-
49
The correctionconfirmed
Participants weren't told their partner might be an AI until afterward
The page says “"Every participant was then told that their partner had been an AI model."”
-
50
The correctionconfirmed
Mistral's new trillion-parameter Large 4
The page says “"a 1 trillion-parameter natively multimodal model with 52 billion active parameters"”
-
51
The correctionconfirmed
Quote "significantly outperforms" any open-weight model built in the US or Europe (2 words, attributed)
The page says “Headline: "says "significantly outperforms" any open-weight models developed in US or Europe"”
-
52
The correctionconfirmed
Tech.eu headline, Oct. 6
The page says “"06 October 2026"”
-
53
The correctionconfirmed
The headline quotes Mistral fairly
The page says “"significantly outperforming any open-weight model developed in the US or Europe"”
-
54
The correctionconfirmed
Its Oct. 6 post
The page says “"October 6, 2026"”
-
55
The correctionconfirmed
Backs the claim with in-house and third-party tests
The page says “Internal human eval vs GLM-5.3, internal STEM eval; third-party Artificial Analysis, vals.ai, Surge AI etc”
-
56
The correctionconfirmed
Against Chinese open-weight models, the post claims to be competitive, ahead on some tests
The page says “"competitive with the strongest open-source models globally"; "leads open-weight models developed outside China by a wide margin"; ahead of Kimi K3 and GLM-5.3 in Surge review. (Page does not itself label specific…”
-
57
The correctionconfirmed
Blind coding review run with Surge AI
The page says “"We also ran a blind human evaluation with Surge AI on coding quality."”
-
58
The correctionconfirmed
Large 4 scored 3.74 out of 5
The page says “ML4 Preview: 3.74 on a 1-5 scale”
- 59 confirmed
-
60
The correctionconfirmed
3.59 for Kimi K3
The page says “Kimi K3: 3.59”
-
61
The correctionconfirmed
4.22 for Claude Opus 5
The page says “Claude Opus 5: 4.22”
- 62 unsupported
-
63
The correctionconfirmed
It's a preview on Mistral's API
The page says “public preview available through the API on Mistral Studio”
-
64
The correctionconfirmed
Weights promised by month's end
The page says “"Weights drop end of this month"”
-
65
The correctionconfirmed
Until the weights ship, outsiders can't test on their own hardware
The page says “Weights not yet released ("Weights drop end of this month"); reasoning follows”
- 66 unsupported
-
67
The correctionconfirmed
Blind coding review with Surge AI, as reported by Mistral
The page says “"blind human evaluation with Surge AI on coding quality"”
- 68 unsupported
-
69
The correctionconfirmed
Opus 5 4.22; Large 4 3.74; GLM-5.3 3.60; Kimi K3 3.59
The page says “Claude Opus 5: 4.22; ML4 Preview: 3.74; GLM-5.3: 3.60; Kimi K3: 3.59”
-
70
The correctionconfirmed
Mistral's own results. Large 4 was tested as a preview.
The page says “Table label "ML4 Preview"; results published in Mistral's post”
-
71
The checklist, item 1confirmed
Pilots loosen oversight step by step
The page says “Stage 1 to 3 progressively reduce physician review”
-
72
The checklist, item 2confirmed
A test of whether people notice an AI is not a test of whether they can spot one
The page says “Generic habit; consistent with 'without being told what it is'”
-
73
The checklist, item 3confirmed
"Best in the US or Europe" is a smaller claim than "best"; open means weights you can download today
The page says “Generic habit; consistent with post. Quoted phrase is a paraphrase, not Mistral's words; see flags”
-
74
art svg/altconfirmed
SVG text 'FINE PRINT'; alt 'A page of text with a pink magnifying glass over its smallest lines, labelled Fine print'
The page says “No factual claim; alt matches drawing (page, pink circle with handle, 'FINE PRINT' label)”
Checked for consistency within the issue itself
The checker's notes
- D
- r
- a
- f
- t
- h
- a
- s
- n
- o
- l
- i
- n
- k
- e
- d
- i
- n
- _
- p
- o
- s
- t
- f
- i
- e
- l
- d
- .
- W
- i
- n
- d
- o
- w
- c
- h
- e
- c
- k
- c
- l
- e
- a
- n
- :
- a
- l
- l
- c
- i
- t
- e
- d
- c
- l
- a
- i
- m
- s
- d
- a
- t
- e
- d
- O
- c
- t
- 1
- -
- 6
- ,
- 2
- 0
- 2
- 6
- (
- i
- n
- s
- i
- d
- e
- 8
- -
- d
- a
- y
- w
- i
- n
- d
- o
- w
- e
- n
- d
- i
- n
- g
- 2
- 0
- 2
- 6
- -
- 1
- 0
- -
- 0
- 7
- )
- ;
- n
- o
- s
- o
- u
- r
- c
- e
- p
- o
- s
- t
- d
- a
- t
- e
- s
- D
- .
- A
- g
- r
- e
- e
- m
- e
- n
- t
- s
- i
- g
- n
- a
- t
- u
- r
- e
- d
- a
- t
- e
- (
- S
- e
- p
- t
- 2
- 1
- -
- 2
- 2
- )
- i
- s
- n
- o
- t
- s
- t
- a
- t
- e
- d
- i
- n
- t
- h
- e
- i
- s
- s
- u
- e
- .
- N
- o
- c
- o
- m
- p
- l
- i
- a
- n
- c
- e
- o
- r
- f
- i
- r
- s
- t
- -
- p
- e
- r
- s
- o
- n
- i
- s
- s
- u
- e
- s
- f
- o
- u
- n
- d
- .