The record for this issue
A new word for the same AI.
71 claims in the draft, each tested by a second agent against the page it cites. 71 held. Nothing had to be cut.
-
01
titleconfirmed
The label changed. The facts didn't.
The page says “Sec. 3(a): 'Super Intelligence' and 'SI' refer to the technologies covered by 'artificial intelligence' as defined in 15 U.S.C. 9401(3)”
-
02
dekconfirmed
A rename became an era
The page says “Fact Sheet: President Donald J. Trump Inaugurates The Era of Super Intelligence ... replacing 'Artificial Intelligence' with 'Super Intelligence'”
-
03
dekconfirmed
a cloud attack where Microsoft saw no ransom note
The page says “we did not observe a ransom note or confirm successful data exfiltration in the activity described here”
-
04
dekconfirmed
became "AI ransomware"
The page says “Headline: AI Ransomware Wiped 100 Azure Accounts in 7 Minutes: Only Pre-Configured Locks Survived”
-
05
dekconfirmed
One claim held up.
The page says “GLM-5.3 develops end-to-end exploits in 50 of 410 attempts ... Claude Mythos Preview did so at a similar rate—in 56 of 410 attempts”
-
06
summaryconfirmed
Three AI claims checked against an executive order, a Microsoft incident report and an Anthropic evaluation
The page says “Executive Order 'Inaugurating The Era Of Super Intelligence'; Microsoft Security blog Storm-3168 report; Anthropic research post on GLM-5.3”
-
07
art svg/altconfirmed
Illustrative tag marked 'NEW LABEL'; no factual claim
The page says “Consistent with the order's renaming of AI to SI”
-
08
The storyconfirmed
In two the label ran ahead of the text; one held up with a caveat
The page says “Mythos Preview did so at a similar rate; caveats: simulated environment, small samples”
-
09
The correctionconfirmed
White House says the President "inaugurates the era of super intelligence."
The page says “Fact Sheet: President Donald J. Trump Inaugurates The Era of Super Intelligence”
-
10
The correctionconfirmed
White House fact sheet headline, Sept. 29
The page says “The White House, September 29, 2026”
-
11
The correctionconfirmed
It's a vocabulary change, and the fact sheet's first point says so
The page says “First bullet: The Order directs all departments and agencies to use the terms 'Super Intelligence' and 'SI' in official correspondence”
-
12
The correctionconfirmed
Agencies must write 'Super Intelligence' and 'SI' for 'AI' in non-statutory documents
The page says “Sec. 2: agencies must use 'Super Intelligence' and 'SI' instead of 'Artificial Intelligence' and 'AI' across official correspondence ... and other non-statutory executive branch documents”
-
13
The correctionconfirmed
where the law allows
The page says “to the maximum extent permitted by law”
-
14
The correctionconfirmed
Section 3 defines the term as technologies covered by the federal AI definition (15 U.S.C. 9401(3))
The page says “Sec. 3(a): refer to the technologies and systems covered by 'artificial intelligence' as defined in 15 U.S.C. 9401(3)”
-
15
The correctionconfirmed
Past rules and contracts don't change
The page says “Sec. 2(b): nothing requires changing previously issued regulations, Presidential actions, contracts, grants, or other historical documents”
-
16
The correctionconfirmed
The administration says the name reflects the technology's transformative capabilities
The page says “Sec. 1: terminology 'should reflect the transformative capabilities of these technologies'”
-
17
The correctionconfirmed
The order sets no capability test
The page says “Sec. 3(a) adopts the existing statutory AI definition; no capability threshold”
-
18
The correctionconfirmed
Bostrom: "greatly exceeds the cognitive performance of humans in virtually all domains of interest" (13 words)
The page says “defines it as 'any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest'”
-
19
The correctionconfirmed
Defense One notes today's systems don't qualify
The page says “today's AI isn't AGI yet, much less ASI”
-
20
The correctionconfirmed
In federal documents, 'SI' means the tools you already use
The page says “Sec. 3(a) ties SI to the existing 15 U.S.C. 9401(3) AI definition”
- 21 confirmed
-
22
The correctionconfirmed
Words in SVG: 'SI' over struck-through 'AI'
The page says “use 'Super Intelligence' and 'SI' in place of 'Artificial Intelligence' and 'AI'”
-
23
The correctionconfirmed
Page where AI is struck through and SI written above
The page says “Sec. 1/2 replacement of terms”
-
24
The correctionconfirmed
Agencies now write 'SI' where they would have written 'AI' in non-statutory documents
The page says “Sec. 2: other non-statutory executive branch documents”
-
25
The correctionconfirmed
past documents don't change
The page says “Sec. 2(b): previously issued regulations ... contracts, grants, or other historical documents”
-
26
The correctionconfirmed
Section 3 points the new term at the existing federal AI definition
The page says “Sec. 3(a) ... 15 U.S.C. 9401(3)”
-
27
The correctionconfirmed
AI ransomware wiped 100 Azure accounts in seven minutes
The page says “Headline: AI Ransomware Wiped 100 Azure Accounts in 7 Minutes”
-
28
The correctionconfirmed
Tech Times headline, Sept. 28
The page says “September 28, 2026, 10:45 AM EDT”
-
29
The correctionconfirmed
Microsoft's Sept. 25 report
The page says “Microsoft Security Blog, September 25, 2026”
-
30
The correctionconfirmed
describes one organization's Azure tenant
The page says “two compromised service principals belonging to the same tenant”
-
31
The correctionconfirmed
The '100' were storage accounts inside it
The page says “The headline's '100 Azure Accounts' refers to these storage accounts”
-
32
The correctionconfirmed
a seven-minute destructive run
The page says “The destructive sequence lasted for about 7 minutes”
-
33
The correctionconfirmed
made 100+ deletion attempts
The page says “100+ storage account deletion attempts”
-
34
The correctionconfirmed
most succeeded
The page says “more than 100 attempts to delete storage accounts, most of which were successful”
-
35
The correctionconfirmed
two compromised service principals
The page says “two compromised service principals belonging to the same tenant”
-
36
The correctionconfirmed
one had its secret exposed in plain text in a public GitHub issue
The page says “client ID, client secret, and tenant ID had previously been exposed in plaintext in a public GitHub issue”
-
37
The correctionconfirmed
Microsoft saw no ransom note
The page says “we did not observe a ransom note”
-
38
The correctionconfirmed
couldn't confirm data theft
The page says “or confirm successful data exfiltration”
-
39
The correctionconfirmed
consistent with tactics that can support ransomware
The page says “consistent with tactics that can support ransomware and extortion operations”
-
40
The correctionconfirmed
Microsoft ties the attack to a shift toward AI-orchestrated attacks
The page says “This activity highlights a broader shift toward AI-orchestrated attacks”
-
41
The correctionconfirmed
Locks set in advance saved a few accounts
The page says “resource locks and storage account-level deletion protection 'blocked deletion attempts for few of the storage accounts'”
-
42
The correctionconfirmed
Rotate any secret that has touched a public repo
The page says “Rotate compromised or exposed credentials immediately ... publicly exposed secrets should be treated as compromised even after removal”
-
43
The correctionconfirmed
lock what you can't lose
The page says “Protect backup and recovery infrastructure as part of ransomware resilience; locks blocked deletions”
-
44
The correctionconfirmed
Two layers from the report
The page says “exposed secret; resource locks/deletion protection”
-
45
The correction, item 1confirmed
Keep secrets out of public repos and issues
The page says “exposed in plaintext in a public GitHub issue; treat publicly exposed secrets as compromised”
-
46
The correction, item 2confirmed
Lock what you can't lose
The page says “Azure resource locks ... blocked deletion attempts”
-
47
The correctionconfirmed
STORAGE
The page says “100+ storage account deletion attempts”
-
48
The correctionconfirmed
The first layer failed: a secret sat in plain text in a public GitHub issue
The page says “exposed in plaintext in a public GitHub issue”
-
49
The correctionconfirmed
The second held where it was set: locks set in advance saved a few accounts
The page says “protections were configured before the attack began; MS: blocked deletion attempts for few of the storage accounts”
-
50
The correctionconfirmed
Z.ai's open-weight GLM-5.3 nearly matches Claude Mythos Preview at building cyber exploits
The page says “Headline: Anthropic says Zhipu's open-weight GLM-5.3 nearly matches Claude Mythos Preview at building exploits”
-
51
The correctionconfirmed
The Decoder headline on Anthropic's Sept. 29 report
The page says “Anthropic post dated September 29, 2026; Decoder published 2026-09-30T11:05:05Z”
-
52
The correctionconfirmed
GLM-5.3 is open-weight and from Z.ai
The page says “GLM-5.3 is released as an open-weight model ... developed by Zhipu AI (known outside of China as Z.ai)”
-
53
The correctionconfirmed
GLM-5.3 built working V8 exploits in 50 of 410 attempts
The page says “GLM-5.3 develops end-to-end exploits in 50 of 410 attempts”
-
54
The correctionconfirmed
against 56 for Mythos Preview
The page says “Claude Mythos Preview did so at a similar rate—in 56 of 410 attempts”
-
55
The correctionconfirmed
binary exploitation: full control-flow hijacks 4% vs 6%
The page says “GLM-5.3 develops full control-flow hijacks in 4% of the trials; Claude Mythos Preview did so in 6%”
-
56
The correctionconfirmed
Both fail most of the time
The page says “50/410 (12%), 56/410 (14%), 4%, 6%”
-
57
The correctionconfirmed
a false cover story got GLM-5.3 to engage 64% of the time
The page says “Providing a deceptive prompt, such as telling the model that it is an autonomous red-team agent ... This gets GLM-5.3 to engage 64% of the time”
-
58
The correctionconfirmed
an abliterated copy, with refusals stripped out, 100%
The page says “Using an abliterated version of the model ... This gets GLM-5.3 to engage 100% of the time”
-
59
The correctionconfirmed
The wider gap was safeguards
The page says “these attacks did not succeed against safeguarded Claude models in our testing; Claude's safeguards blocked the requests that used deceptive prompts”
-
60
The correctionconfirmed
Refusals can be stripped from open weights
The page says “users can reconfigure it to remove its refusals with little change in its capabilities”
-
61
The correctionconfirmed
security teams should assume attackers have this level of help
The page says “Cyber defenders face attackers who will use every capable tool they can”
-
62
The correctionconfirmed
the test is Anthropic's, run on a rival's model
The page says “Anthropic's own evaluations of Zhipu's GLM-5.3”
-
63
The correctionconfirmed
Close on capability. GLM-5.3's guardrails gave way.
The page says “similar rate ... released without meaningful safeguards to limit misuse”
-
64
The correctionconfirmed
Anthropic's tests of Z.ai's GLM-5.3
The page says “developed by Zhipu AI (known outside of China as Z.ai)”
-
65
The correctionconfirmed
V8 exploits share of 410 attempts: 12% vs 14%
The page says “50 of 410 (12.2%) and 56 of 410 (13.7%), rounded”
-
66
The correctionconfirmed
Control-flow hijacks: 4% vs 6%
The page says “4% of the trials; ... 6%”
-
67
The correctionconfirmed
Engagement: false cover story 64%, abliterated 100%
The page says “engage 64% of the time ... engage 100% of the time”
-
68
The correctionconfirmed
Both models fail most exploit attempts. Stripping refusals took engagement to 100%.
The page says “50/410, 56/410; abliterated version engages 100% of the time”
-
69
The checklist, item 1confirmed
Find the document; definitions live in the text
The page says “Definition is in Sec. 3 of the order”
-
70
The checklist, item 2confirmed
'100 accounts' can mean one organization's storage
The page says “affected one organization ... '100 Azure Accounts' refers to these storage accounts”
-
71
The checklist, item 3confirmed
'Nearly matches' can mean 12% against 14%
The page says “50 of 410 vs 56 of 410”
The checker's notes
- Defense One article is dated Sept. 22, 2026, exactly 8 days before D, inside the 8-day window. It predates the order and covers the UN speech; the draft cites it only for the definition and capability point, which it supports.
- White House order page shows a later modified date (Oct. 2, 2026); the order itself was signed and published Sept. 29, 2026.
- No compliance issues found: no market, valuation or investing content; no first-person opinion; all quotes 14 words or fewer and attributed.