The record for this issue
The careful words got dropped.
79 claims in the draft, each tested by a second agent against the page it cites. 77 held; 2 did not and were fixed or cut before this went out.
-
01
titleconfirmed
The careful words got dropped.
The page says “Fair summary of three corrections below, each confirmed against its source (editorial headline)”
-
02
dekconfirmed
Postgres's official reason for its delay never mentions AI.
The page says “"Given the volume of fixes that have been added to the back branches..." (no mention of AI, LLM or automated review)”
-
03
dekconfirmed
A hacker's prompts were in Chinese
The page says “AIA0001: "This activity cluster uses Chinese language." Reuters/CP24: instructions in "fewer than 100 Chinese characters"”
-
04
dekconfirmed
the report names no nationality
The page says “Report labels clusters by language (AIA0001 Chinese, AIA0002 English); states no operator nationality”
-
05
dekconfirmed
Anthropic staff said AI could do most of their work, not that it will
The page says “"Respondents were asked when AI will be able to complete these tasks, regardless of whether companies will actually use it."”
-
06
summaryconfirmed
Three AI claims from the past week: PostgreSQL 19 delay, AI-agent hacking campaign, Anthropic staff survey
The page says “Claims dated Oct 16 (The Register), Oct 20 (TechRadar), Oct 17 (NeoTeo), all within 8 days of Oct 21”
-
07
art alt / art svg textconfirmed
Missing word 'Could' (SVG text: COULD)
The page says “Consistent with correction 3 (capability question, 'could'); no other words in SVG”
-
08
The storyconfirmed
The hedge was in the source. The headline left it out.
The page says “Editorial summary consistent with the three corrections”
-
09
The storyconfirmed
Sources are a release team's email, a security firm's report and a survey write-up
The page says “Release team email (Jonathan Katz, pgsql-hackers, Oct 10); NorForge research post; Anthropic research write-up”
-
10
The storyconfirmed
Each headline dropped that care
The page says “Register/TechRadar/NeoTeo headlines as quoted in the corrections”
-
11
The correctionconfirmed
PostgreSQL 19 was delayed because AI reviewers found more bugs than humans can fix
The page says “Headline: "Postgres 19 delayed after AI reviewers find more bugs than humans can fix"”
-
12
The correctionconfirmed
The Register headline, Oct. 16
The page says “Published Friday, 16 October 2026”
-
13
The correctionconfirmed
On Oct. 10 the release team moved GA to Nov. 19
The page says “Date 2026-10-10; "The release team decided to move the PostgreSQL 19 GA to November 19"”
-
14
The correctionconfirmed
citing "the volume of fixes"
The page says “"Given the volume of fixes that have been added to the back branches"”
-
15
The correctionconfirmed
fixes added to older branches
The page says “"added to the back branches"”
-
16
The correctionconfirmed
wants them in a second release candidate on Nov. 12
The page says “"make those available in a PostgreSQL 19 RC2 that would be scheduled for November 12"”
-
17
The correctionconfirmed
Its email doesn't mention AI
The page says “Full body contains no reference to AI, LLMs or automated review”
-
18
The correctionconfirmed
committer Peter Eisentraut
The page says “"as the developer or committer of some of those features"”
-
19
The correctionconfirmed
personal Oct. 9 blog post
The page says “Dated Oct 9, 2026; "these are just my opinions at this point"; "from my personal perspective"”
-
20
The correctionconfirmed
LLM-assisted review finds problems faster
The page says “"with LLM-assisted code review, these kinds of things can get found much faster"”
-
21
The correctionconfirmed
raising the bar for older code
The page says “"the effective quality bar has been raised after the time the feature was written and committed"”
-
22
The correctionconfirmed
about eight to 10 significant features were pulled
The page says “"about 8 to 10 significant features, depending on how you count, have been reverted"”
-
23
The correctionconfirmed
with more time, they could have been fixed
The page says “"with more time, say, two months, these could have been addressed"”
-
24
The correctionwrong
The headline's "can't fix"
The page says “Headline: "Postgres 19 delayed after AI reviewers find more bugs than humans can fix"”
-
25
The correctionconfirmed
AI review produces findings faster than people can work through them
The page says “"these kinds of things can get found much faster"; each fix has overhead and the team "run[s] out of time"”
-
26
The correctionconfirmed
Missing context
The page says “Official reason is volume of back-branch fixes; AI link is one contributor's personal post; 'more time' could have fixed them”
-
27
The correctionconfirmed
How the release moved; 2026
The page says “All dates in 2026”
-
28
The correction, step 1confirmed
Oct. 9: committer's blog ties feature reverts to faster, LLM-assisted review
The page says “Oct 9, 2026; "with LLM-assisted code review, these kinds of things can get found much faster"”
-
29
The correction, step 2confirmed
Oct. 10: release team moves GA, citing the volume of fixes
The page says “2026-10-10; "Given the volume of fixes..."”
-
30
The correction, step 3confirmed
Oct. 15: first release candidate ships
The page says “"October 15, 2026: PostgreSQL 19 RC 1 Released!"”
-
31
The correction, step 4confirmed
Nov. 12: second release candidate planned
The page says “"The release team plans to issue a second release candidate, PostgreSQL 19 RC 2, on November 12, 2026."”
-
32
The correction, step 5confirmed
Nov. 19: general availability planned
The page says “"The planned date for the general availability of PostgreSQL 19 is November 19, 2026."”
-
33
The correctionconfirmed
Email gives volume of fixes as reason; AI link only from Oct. 9 blog post
The page says “Email and RC1 announcement both omit AI; the AI link appears in Eisentraut's post (and The Register's report of it)”
-
34
The correctionconfirmed
A Chinese hacker used AI agents to break into more than 100 company websites
The page says “Headline: "Chinese hacker hits over 100 company websites using AI agents"”
-
35
The correctionconfirmed
TechRadar headline, Oct. 20
The page says “Published 20 October 2026”
-
36
The correctionconfirmed
NorForge's Oct. 19 report
The page says “Byline "OCT 19, 2026"”
-
37
The correctionconfirmed
The research firm NorForge
The page says “Page describes its work as "NorForge research" (tagline "Forging trust in autonomous systems")”
-
38
The correctionconfirmed
labels one cluster by its Chinese-language instructions
The page says “AIA0001: "This activity cluster uses Chinese language."”
-
39
The correctionconfirmed
aimed at online shops
The page says “AIA0001 targets "e-commerce/payment infrastructure"”
-
40
The correctionconfirmed
in the US, Japan and Australia
The page says “"mostly in the US, Japan and Australia" (task text also lists Canada)”
-
41
The correctionconfirmed
It names no nationality
The page says “No operator nationality stated; clusters labelled by language”
-
42
The correctionconfirmed
calls the activity financially motivated
The page says “"all appeared to have financial motivation"”
-
43
The correctionconfirmed
with no obvious state-sponsored intrusions
The page says “"there was no obvious state-sponsored intrusion activity"”
-
44
The correctionconfirmed
count of more than 100 companies comes from Reuters' review of the attacker's activity
The page says “"more than 100 companies"; Reuters "was able to establish the victims' identities by independently reviewing the hacker's Cairn activity"”
-
45
The correctionconfirmed
report itself gives case studies, not a total
The page says “Six case studies; counts of panels/projects, no victim total”
-
46
The correctionconfirmed
success rate around 10%
The page says “"average success rate is hovering around 10%"”
-
47
The correctionconfirmed
warns its cost estimates could be off by a clear margin
The page says “"The uncertainty here is however considerable, and numbers could be off by a clear margin." (re: cost guesstimates)”
-
48
The correctionconfirmed
when a company's main site held, the agents looked for its side doors
The page says “"After failing to find weaknesses in a large company's public-facing application, the AI agents pored through digital records to find associated infrastructure"”
-
49
The correctionconfirmed
Overstated
The page says “Nationality not stated in primary report; 100+ count from Reuters, not the report”
-
50
The correctionconfirmed
How the agents got in; as described in Reuters' account
The page says “Reuters exclusive via CP24, Oct 19, 2026”
-
51
The correction, step 1confirmed
Probe the company's public app
The page says “"failing to find weaknesses in a large company's public-facing application"”
-
52
The correction, step 2confirmed
When it holds, map related systems
The page says “"pored through digital records to find associated infrastructure"”
-
53
The correction, step 3confirmed
Find a public-facing control panel
The page says “"located a public-facing control panel"”
-
54
The correction, step 4confirmed
Use a data-processing flaw to run commands
The page says “"a subtle weakness in the way it processed data"; exploited to make the server run malicious commands”
-
55
The correctionconfirmed
One victim said the hit was on a standalone support site, and was contained
The page says “SharkNinja: "the hack had hit a standalone site carrying only product support information and that the threat had been contained"”
-
56
The correctionconfirmed
Anthropic employees expect AI to do 75% of their work by 2030
The page says “Headline: "Anthropic employees expect AI to do 75% of their work by 2030"”
-
57
The correctionconfirmed
NeoTeo headline, Oct. 17
The page says “Published October 17, 2026”
-
58
The correctionconfirmed
75% is the average staff answer
The page says “Staff expect AI "to be able to do 75% of their current work by 2030"”
-
59
The correctionconfirmed
1,334 staff
The page says “"an internal poll that ran from August 9 to 12 and drew 1,334 respondents"”
-
60
The correctionconfirmed
unweighted internal poll
The page says “staff responses are "unweighted"”
-
61
The correctionconfirmed
Aug. 9-12
The page says “"ran from August 9 to 12"”
-
62
The correctionconfirmed
asked when AI could do their tasks
The page says “"Respondents were asked when AI will be able to complete these tasks"”
-
63
The correctionconfirmed
quote: "regardless of whether companies will actually use it."
The page says “"...regardless of whether companies will actually use it." (8 words, attributed to the survey question)”
-
64
The correctionconfirmed
weighted Morning Consult survey
The page says “responses "are weighted to be representative of the adult population"”
-
65
The correctionconfirmed
10,980 US adults
The page says “10,980 US adults, August 11-23 survey”
-
66
The correctionconfirmed
those with jobs averaged 28%
The page says “"US workers (respondents who report having a job) put the share at 28% on average"”
-
67
The correctionconfirmed
Staff used a slider; the public picked ranges
The page says “"Staff answered on a slider while US adults choose a range and we use range midpoints."”
-
68
The correctionconfirmed
Anthropic says these are personal estimates, not company forecasts or plans
The page says “"the personal expectations of staff members in an internal poll"; "do not represent Anthropic's plans or its view of its headcount"; "not a measurement by Anthropic or a company forecast"”
-
69
The correctionconfirmed
measures belief in capability, not a plan to cut jobs
The page says “"do not represent Anthropic's plans or its view of its headcount"”
-
70
The correctionconfirmed
over half of staff expected AI to do at least 80% of their work
The page says “Over half of staff expect AI "to be able to do at least 80% of their current work by 2030"”
-
71
The correctionconfirmed
against 5% of US workers
The page says “"only 1 in 20 say that AI could do 80% or more"”
-
72
The correctionunsupported
The people building the tools expect far more of them than the people using them.
-
73
The correctionconfirmed
Right, but
The page says “75% figure correct; question was about capability, not use or plans”
-
74
The correctionconfirmed
Two groups, one question, two formats
The page says “Figure 1 caption covers both groups; "Staff answered on a slider while US adults choose a range"”
-
75
The correctionconfirmed
Average share of their own work people think AI could do by 2030
The page says “75% / 28% averages, 2030”
-
76
The correctionconfirmed
Share of current work
The page says “"75% of their current work"”
-
77
The correctionconfirmed
Anthropic staff 75
The page says “"75% of their current work by 2030"”
-
78
The correctionconfirmed
Employed US adults 28
The page says “"US workers (respondents who report having a job) put the share at 28% on average"”
-
79
The correctionconfirmed
Staff on a slider; US adults chose a range scored at its midpoint
The page says “"Staff answered on a slider while US adults choose a range and we use range midpoints."”
The checker's notes
- Checklist items (blocks[4]) are reusable reading habits with no checkable factual claim; not counted.
- No securities/market content, no first-person statements, all quotes 14 words or fewer and attributed.
- No cited source is dated after D; Nov. 12/19 dates were announced Oct. 10.