The record for this issue
The agent got a badge.
105 claims in the draft, each tested by a second agent against the page it cites. 100 held; 5 did not and were fixed or cut before this went out.
-
01
titleconfirmed
The agent got a badge. (agents get own work identity)
The page says “The agent receives its own Workspace account, including an email address, calendar, Drive, and presence in your company directory”
-
02
dekconfirmed
Google's coworker agents get their own work account, email address and place in the company directory
The page says “The agent receives its own Workspace account, including an email address, calendar, Drive, and presence in your company directory”
-
03
dekconfirmed
every action is logged under the agent's name
The page says “every action Gemini takes is written to an audit trail and attributed to the agent rather than to a person”
-
04
dekconfirmed
The same week, OpenAI and Google made AI-made content easier to detect
The page says “Oct. 5: 'Starting today, API customers around the world will be able to opt in to watermarked text outputs'; Google SynthID Detector opened to everyone Oct. 7 (blog.google)”
-
05
summaryconfirmed
Google gave Gemini coworker agents their own Workspace accounts and audit trails
The page says “The agent receives its own Workspace account ... every action Gemini takes is written to an audit trail and attributed to the agent”
-
06
summarywrong
OpenAI began watermarking text in the EU
The page says “Over the coming weeks, we will add an invisible watermark to eligible ChatGPT and Codex text output in the European Union”
-
07
summaryconfirmed
SynthID Detector opened to the public
The page says “Now, we're expanding access to everyone – with the tool available globally in English starting today”
-
08
art alt / art svgconfirmed
Badge labelled AGENT (illustration, no factual content beyond the lead story)
The page says “A coworker agent acts under its own identity rather than yours”
-
09
The week in four, item 1confirmed
Google's coworker agents get their own Workspace account, email address and directory listing
The page says “its own Workspace account, including an email address, calendar, Drive, and presence in your company directory”
-
10
The week in four, item 1confirmed
act under their own identity, not yours
The page says “A coworker agent acts under its own identity rather than yours”
-
11
The week in four, item 2confirmed
OpenAI will watermark ChatGPT and Codex text for eligible EU users over the coming weeks
The page says “Over the coming weeks, we will add an invisible watermark to eligible ChatGPT and Codex text output in the European Union”
-
12
The week in four, item 2confirmed
to meet the EU AI Act
The page says “The EU AI Act requires generative AI providers to make generated text identifiable in a machine-readable way”
-
13
The week in four, item 3confirmed
Google opened SynthID Detector to the public for images, video and audio
The page says “anyone can easily check if an image, video, or audio file was made with AI from Google or our partners”
-
14
The week in four, item 3confirmed
including watermarks from OpenAI, Nvidia and Kakao
The page says “including OpenAI, NVIDIA, Kakao, and soon, Apple”
-
15
The week in four, item 4confirmed
GPT-6 brought interactive charts, forms and tools to ChatGPT answers
The page says “responses can include graphics, tappable buttons, forms, charts, and interactive experiences”
-
16
The week in four, item 4confirmed
Anthropic put Claude inside Google Docs, Sheets and Slides
The page says “A sidebar opens up alongside your file in Docs, Sheets, or Slides”
-
17
The storyconfirmed
Gemini at Work event
The page says “This article is adapted from Thomas Kurian's keynote address at Gemini at Work 2026”
-
18
The storyconfirmed
on Thursday, Oct. 8
The page says “At a Google Cloud event on Thursday (article dated October 8, 2026)”
-
19
The storyconfirmed
a single 'universal agent for work' in Gemini
The page says “the Gemini agent: your new single, universal agent for work”
-
20
The storyconfirmed
coworker agents you create by describing a role
The page says “You simply describe the role you need, and Gemini creates it”
-
21
The storyconfirmed
own Workspace account: an email address, a calendar, Drive and a place in the company directory
The page says “its own Workspace account, including an email address, calendar, Drive, and presence in your company directory”
-
22
The storyconfirmed
quote 'acts under its own identity rather than yours' and sees only what you share
The page says “A coworker agent acts under its own identity rather than yours, and it sees only what you share with it”
-
23
The storyconfirmed
Businesses get it first
The page says “Google will initially focus on bringing the agent to businesses before later rolling it out to consumers”
-
24
The storyconfirmed
Google gave no general-availability date or price
The page says “Neither the Google Cloud post nor TechCrunch gives a GA date or price for the Gemini agent (only industry previews and spend-cap controls)”
-
25
The storyconfirmed
every action written to an audit trail attributed to the agent rather than a person
The page says “every action Gemini takes is written to an audit trail and attributed to the agent rather than to a person”
-
26
The storyconfirmed
permissions approved by the organization's security administrators
The page says “role-based access permissions, approved by your organization's security administrators”
-
27
The storyconfirmed
it runs in a sandbox
The page says “All Gemini agents execute tasks safely inside an Agent Sandbox with its own network boundary”
-
28
The storyconfirmed
traffic passes through a gateway that enforces policy in real time
The page says “passes through Agent Gateway, an AI network firewall enforcing your organization's policies in real time”
-
29
The storyconfirmed
When a project hits its spend cap, that project's agent pauses until someone resumes it
The page says “if a spend cap is triggered, that project's agent pauses, and you can choose to resume work with a single click in the console”
-
30
The storyconfirmed
On Sept. 25, Microsoft said Copilot's Autopilot agent keeps a distinct identity, memory and workspace
The page says “the system maintains a distinct identity, memory, computing environment and workspace (announced Friday, Sept. 25, 2026)”
-
31
The storyconfirmed
On Sept. 23, Meta said its Muse agent will get its own email address
The page says “Muse will get its own email address (TechCrunch, Sept. 23, 2026)”
-
32
The storyconfirmed
Google is the third large vendor in a month to treat an agent as a colleague with an identity
The page says “Follows from Sept. 23 (Meta), Sept. 25 (Microsoft), Oct. 8 (Google) dates on the three cited pages; 'third' is the issue's count of these three”
-
33
The storyconfirmed
an agent with its own account can be scoped, audited and paused
The page says “fine-grained, role-based access permissions ... audit trail ... that project's agent pauses”
-
34
The storyunsupported
It also needs ... a person who owns it, and someone who closes it when the work is done
-
35
The storyconfirmed
Agents with an identity of their own; announcement dates
The page says “Dates match the three cited pages”
-
36
The story, step 1confirmed
Sept. 23 Meta: Muse will get its own email address
The page says “Muse will get its own email address”
-
37
The story, step 2confirmed
Sept. 25 Microsoft: Autopilot keeps a distinct identity and workspace
The page says “maintains a distinct identity, memory, computing environment and workspace”
-
38
The story, step 3confirmed
Oct. 8 Google: coworker agents get a Workspace account and directory listing
The page says “own Workspace account ... presence in your company directory (post dated October 8, 2026)”
-
39
The story, item 1confirmed
Audit trail in the agent's own name
The page says “attributed to the agent rather than to a person”
-
40
The story, item 2confirmed
A spend cap that pauses it
The page says “if a spend cap is triggered, that project's agent pauses”
-
41
The story, item 3confirmed
A gateway that enforces policy
The page says “Agent Gateway, an AI network firewall enforcing your organization's policies in real time”
-
42
The story, item 4confirmed
A sandbox it runs in
The page says “inside an Agent Sandbox with its own network boundary”
-
43
The story, item 5confirmed
Permissions your admins approve
The page says “approved by your organization's security administrators”
-
44
The story, item 6confirmed
Only what you share with it
The page says “it sees only what you share with it”
-
45
The storyunsupported
None of these layers names a person who owns the agent; deciding who answers for it is up to the company
-
46
The storyconfirmed
The controls Google described
The page says “All six layers appear in the Google Cloud post”
-
47
shipped, item 1confirmed
OpenAI GPT-6 with Intelligent UI, Oct. 7
The page says “GPT-6 in ChatGPT introduces Intelligent UI (dated October 7, 2026)”
-
48
shipped, item 1confirmed
ChatGPT can answer with interactive graphics, buttons, forms and charts
The page says “responses can include graphics, tappable buttons, forms, charts, and interactive experiences”
-
49
shipped, item 1confirmed
Rolled out to Plus, Pro, Business and Enterprise on Oct. 7
The page says “starts rolling out globally to ChatGPT Plus, Pro, Business and Enterprise tiers today”
-
50
shipped, item 1confirmed
Enterprise depends on admin settings
The page says “Enterprise availability depends on workplace admin settings”
-
51
shipped, item 1confirmed
to Free and Go on Oct. 8
The page says “Starting tomorrow, the rollout expands to Free and Go tiers. (post dated Oct. 7)”
-
52
shipped, item 1confirmed
Work and Codex models are unchanged
The page says “The models powering Work and Codex are not changing as part of this release”
-
53
shipped, item 2confirmed
Anthropic Claude Haiku 5.5, Oct. 7
The page says “Page dated October 7, 2026”
-
54
shipped, item 2confirmed
$0.10 per million input and $0.50 output for prompts up to 100k tokens
The page says “Haiku 5.5 (prompts up to 100k / over 100k): input $0.10 / $0.50, output $0.50 / $2.50”
-
55
shipped, item 2confirmed
costs around 75% less to run than Haiku 4.5 on average
The page says “On average, it now costs around 75% less to run”
-
56
shipped, item 2confirmed
Sonnet 5.5's cache reads halved, to $0.10 per million
The page says “Cache reads now cost 50% less: $0.10 per million tokens rather than $0.20”
-
57
shipped, item 3confirmed
Claude for Google Workspace, Oct. 6
The page says “public beta starting today (article dated October 6, 2026)”
-
58
shipped, item 3confirmed
sidebar add-on in public beta for all paid Claude users
The page says “The Claude for Google Workspace beta is available to all paid Claude users”
-
59
shipped, item 3confirmed
rewrites sections in Docs
The page says “rewrite portions of a document, or restyle sections”
-
60
shipped, item 3confirmed
writes formulas in Sheets
The page says “In Sheets, you can have Claude write formulas or build tables”
-
61
shipped, item 3confirmed
builds slides from your existing layouts
The page says “Claude can build new slides using existing layouts or themes”
-
62
shipped, item 3confirmed
'ask before edits' mode shows changes before they're applied
The page says “There's an 'ask before edits' mode where Claude will let you preview changes before they're made”
-
63
shippedconfirmed
Answers you can click, and a cheaper small model
The page says “tappable buttons, forms ... ; Haiku 5.5 priced 90% lower than Haiku 4.5 (anthropic.com)”
-
64
shippedconfirmed
Input: Haiku 5.5 $0.10, Haiku 4.5 $1.00
The page says “Haiku 5.5 input $0.10 (≤100k); Haiku 4.5 $1.00”
-
65
shippedconfirmed
Output: Haiku 5.5 $0.50, Haiku 4.5 $5.00
The page says “Haiku 5.5 output $0.50 (≤100k); Haiku 4.5 $5.00”
-
66
shippedconfirmed
List prices for prompts up to 100k tokens
The page says “priced 90% lower than Claude Haiku 4.5 for requests up to 100,000 tokens”
-
67
shippedconfirmed
average saving around 75%, after a new tokenizer that uses slightly more tokens
The page says “On average, it now costs around 75% less to run ... updated tokenizer ... uses slightly more tokens per task”
-
68
shippedunsupported
Small models are now cheap enough to sit behind every routine step of an agent
-
69
The storyconfirmed
On Oct. 5, OpenAI said it will add textGrain ... to ChatGPT and Codex text for eligible EU users over the coming weeks
The page says “Over the coming weeks, we will add an invisible watermark to eligible ChatGPT and Codex text output in the European Union. (Oct. 5, 2026)”
-
70
The storyconfirmed
textGrain is an invisible statistical signal in the model's word choices
The page says “textGrain, adds an invisible statistical signal to the model's word choices”
-
71
The storyconfirmed
EU AI Act requires providers to make generated text identifiable by machine
The page says “requires generative AI providers to make generated text identifiable in a machine-readable way”
-
72
The storyconfirmed
TNW reports existing providers have until Dec. 2 to comply
The page says “Providers already on the market have until 2 December to comply (citing Engadget)”
-
73
The storyconfirmed
API customers anywhere can opt in
The page says “API customers around the world will be able to opt in to watermarked text outputs for select models”
-
74
The storyconfirmed
detector goes only to vetted researchers and organizations, case by case
The page says “Access will initially be limited to approved researchers and expert organizations ... granted on a case-by-case basis. (note: 'initially')”
-
75
The storyconfirmed
At 1% false-positive target, detected about 80% of 200-token passages
The page says “our detector identified watermarks in about 80% of 200-token passages”
-
76
The storyconfirmed
about 95% of 400-token ones
The page says “compared with about 95% of 400-token passages”
-
77
The storyconfirmed
Swapping 10% of words for synonyms cut detection to 66%
The page says “replacing 10% of words with synonyms reduced detection from about 92% to 66%”
-
78
The storyconfirmed
swapping a quarter cut it to 17%
The page says “Replacing 25% of words reduced it to 17%”
-
79
The storyconfirmed
quote 'the absence of a detected watermark does not prove human authorship' (OpenAI, 10 words)
The page says “The absence of a detected watermark does not prove human authorship”
-
80
The storyconfirmed
on Oct. 7 Google opened SynthID Detector to anyone for images, video and audio
The page says “expanding access to everyone ... starting today (Oct. 7, 2026); image, video, or audio file”
-
81
The storyconfirmed
A watermark, and what it can't prove
The page says “The absence of a detected watermark does not prove human authorship”
-
82
The storyconfirmed
OpenAI's tests, at a 1% false-positive target
The page says “at a 1% false-positive rate”
-
83
The storyconfirmed
Unedited passages: 200 tokens 80%, 400 tokens 95%
The page says “about 80% of 200-token passages ... about 95% of 400-token passages”
-
84
The storyconfirmed
400-token passages, words swapped: 10% -> 66%, 25% -> 17%
The page says “In an evaluation of 400-token passages, replacing 10% ... to 66% ... Replacing 25% of words reduced it to 17%”
-
85
The storyconfirmed
detection is substantially lower for math
The page says “Detection rates were substantially lower for content such as mathematics”
-
86
Agent actions and AI output got easier to traceconfirmed
Agents now act under their own names in a log, and AI text, images and audio carry marks a machine can check
The page says “audit trail attributed to the agent; textGrain watermark (openai.com); SynthID Detector for image, video, audio (blog.google)”
-
87
What to watch, item 1confirmed
Wed 10/14: GPT-5.5 leaves Codex for ChatGPT sign-in users at noon Pacific; API-key sessions unaffected
The page says “Effective October 14, 2026, at noon PDT (19:00 UTC), GPT-5.5 retires from Codex for users signed in with ChatGPT. The OpenAI API and Codex sessions authenticated with an API key aren't affected”
-
88
What to watch, item 2confirmed
Mon 10/19: Scheduled Release Workspace domains get Gemini skills from this date
The page says “Scheduled Release domains: Gradual rollout starting on October 19, 2026 (post dated Sept. 30, 2026)”
-
89
What to watch, item 3confirmed
Coming weeks: textGrain reaches eligible ChatGPT and Codex text
The page says “Over the coming weeks, we will add an invisible watermark to eligible ChatGPT and Codex text output”
-
90
What to watch, item 4confirmed
Mistral will release Mistral Large 4's weights by end of month
The page says “We will release the weights by the end of the month. (post dated Oct. 6, 2026)”
-
91
What to watch, item 4confirmed
API preview opened Oct. 6
The page says “You can try the preview API today on Mistral Studio. (Oct. 6, 2026)”
-
92
The checklist, item 2confirmed
If it sees only what you share (premise)
The page says “it sees only what you share with it”
-
93
The checklist, item 3confirmed
Set a spend cap and decide who's allowed to resume it (premise: caps pause the agent until resumed)
The page says “that project's agent pauses, and you can choose to resume work with a single click”
-
94
The checklist, item 4confirmed
No mark doesn't mean a person wrote it
The page says “The absence of a detected watermark does not prove human authorship”
-
95
The checklist, item 4wrong
light edits hide it
The page says “replacing 10% of words with synonyms reduced detection from about 92% to 66%”
-
96
quoteconfirmed
'A coworker agent acts under its own identity rather than yours.' (11 words) — Thomas Kurian, CEO, Google Cloud, via Google Cloud blog
The page says “Byline Thomas Kurian, CEO, Google Cloud; 'This article is adapted from Thomas Kurian's keynote'; 'A coworker agent acts under its own identity rather than yours'”
-
97
Before you go, item 1confirmed
premise: coworker agents get an email address and a calendar
The page says “including an email address, calendar, Drive”
-
98
Before you go, item 2confirmed
premise: OpenAI's watermark lives in word choice
The page says “adds an invisible statistical signal to the model's word choices”
-
99
Before you go, item 3confirmed
premise: ChatGPT now answers with forms
The page says “responses can include graphics, tappable buttons, forms, charts”
-
100
linkedin postconfirmed
Gemini coworker agents each get an email address, a calendar, Drive and a place in the company directory
The page says “its own Workspace account, including an email address, calendar, Drive, and presence in your company directory”
-
101
linkedin postconfirmed
act under their own identity, see only what you share, every action in an audit trail under the agent's name
The page says “acts under its own identity rather than yours, and it sees only what you share with it ... attributed to the agent”
-
102
linkedin postconfirmed
Microsoft and Meta said similar things in September
The page says “Microsoft Sept. 25 (VentureBeat); Meta Sept. 23 (TechCrunch)”
-
103
linkedin postconfirmed
OpenAI said it will watermark ChatGPT text in the EU
The page says “Over the coming weeks, we will add an invisible watermark to eligible ChatGPT and Codex text output in the European Union”
-
104
linkedin postconfirmed
swap a quarter of the words and detection drops to 17%
The page says “Replacing 25% of words reduced it to 17%”
-
105
linkedin postconfirmed
plus a much cheaper small Claude model
The page says “priced 90% lower than Claude Haiku 4.5 for requests up to 100,000 tokens”