IMPLICATOR​.ai
Morning Briefing · From San Francisco

 

Wednesday, October 7, 2026
9 stops

From San Francisco

1
The Editorial
 

Good morning.

AI companies are putting a price on access.

Google signs a 20-year agreement with Constellation for 890 megawatts of new nuclear capacity from existing reactors. The power goes to a grid shared with other customers.

Anthropic opens its Cyber Verification Program to vetted security teams. Its Glasswing partners found at least 129,000 verified software vulnerabilities between April and July 2026. Getting past Claude Mythos's cyber blocks now takes approval.

OpenAI plans two ChatGPT ad formats for free users later this month. An ad sits beneath the progress bar of an image still being made, so the sales pitch gets there first.

Stay curious,

Marcus Schuler

Good briefing? Pass it on
X · LinkedIn · Bluesky · Email
PRO for $59 through Friday
Get your first year of Implicator PRO for $59 instead of $89. The founding offer ends Friday, October 9, at midnight Pacific.
Get PRO for $59 →
First year only. After October 9, PRO moves to $99 a year or $10 a month.
228
prompts
19
categories
456
test runs
The Implicator Prompt Library
Inspect 228 prompts across 19 categories, with 456 test runs on ChatGPT and Claude. One category is open to everyone and more come free with signup. PRO includes everything.
Open the library →
The Paperclip Compendium
Free for newsletter subscribers. We ran Paperclip on our server: agents get an org chart, a task board and budgets. Our guide covers setup, your first agent team, cost control and 23 troubleshooting entries.
Read the compendium →
 
2
The Big Story
Google signs a 20-year deal for 890 MW of new nuclear capacity.

Google signed a 20-year power purchase agreement with Constellation to fund 890 MW of new nuclear capacity for the PJM grid.

The added power comes from raising output at 11 existing Constellation nuclear units in Illinois, Pennsylvania and New Jersey. Constellation commits $4.3 billion in new investment, and first capacity is expected by 2028.

A separate 15-year agreement covers 2,700 MW of energy supply. A five-year technology alliance puts Google Cloud and Gemini to work inside Constellation's operations.

Why This Matters:

Reality Check
What's confirmed: Constellation makes the $4.3 billion investment in the reactor uprates. Google's payment terms are not disclosed.
What's implied (not proven): Constellation's use of Google's AI could return some of the money to Google.
What could go wrong: The uprate schedule could slip. First added capacity is expected by 2028.
What to watch next: Watch for disclosure of the price per megawatt-hour. The release does not specify how much capacity will power Google's own data centers.
Read the announcement →
 
3
Also Today
Anthropic opens vetted access to Claude Mythos for security teams.

Anthropic launched its Cyber Verification Program, giving verified security teams access to advanced Claude Mythos models with fewer cyber blocks.

Project Glasswing partners found at least 129,000 verified software vulnerabilities between April and July 2026. Access to Anthropic's most capable defender models now depends on passing the vendor's checks. Organizations apply through a portal on Anthropic's site.

Read the announcement →
 
4
The Outside Read
 

The Wikimedia Foundation's October 5, 2026, account documents OpenAI agent activity on the sites it maintains.

The account describes cleanup work imposed on a nonprofit run by volunteers, including apparent attempts to misuse a citation tool as a proxy. Its investigation found no evidence of compromise.

Read it at the Wikimedia Foundation →
 
5
The One Number
 
$50B
Moonshot AI's valuation after its latest private round, up from $31.5 billion earlier in summer 2026. That is about 25 times its expected December revenue run rate of $2 billion, versus about $1 billion now. Investors are also paying for its confidentially filed Hong Kong IPO, targeted for the first quarter of 2027.
Source: CTech by Calcalist, October 2026
 
6
Today's Headlines
 
The Next 72 Hours
Wed 10/7
Fed: the central bank publishes the minutes of its September 15-16 meeting at 2 p.m. Eastern.
Wed 10/7
Tech: the Motek automation and production-technology fair continues in Stuttgart, 9 a.m. to 5 p.m. local time.
Thu 10/8
Tech: the IAPP privacy, security, risk and AI governance conference opens in Seattle and runs through Oct. 9.
Tuesdays go deeper.  Sign up for Implicator PRO for the weekly Tuesday deep dive on deploying AI where it pays. $8 a month, $89 a year.
 
7
The 5-Minute Skill
 

A sales deck may promise your data will not train a model while the contract grants broader rights. Test the promise before approval.

Your raw input:

Paste the vendor’s terms of service, privacy notice, data processing addendum, relevant sales claims, and a short list of the data your team would upload.

The prompt:

You are reviewing an AI vendor’s data rights. Write a decision note covering permitted uses, retention and deletion, sharing and subprocessors, and unresolved questions. For every conclusion, quote the shortest controlling phrase and name its source and section. Distinguish customer content, account data, usage data, feedback, and de-identified data. State whether each category may train or improve models. If documents conflict, show both passages and identify which appears to govern. Do not infer consent from silence. End with five vendor questions and a one-sentence go, pause, or reject recommendation based only on the supplied text.

Why this works: The prompt forces the model to tie each claim to contract language instead of producing a generic summary. Separating data categories exposes exceptions hidden behind broad assurances.

What to use: Use the strongest long-context model available in ChatGPT or Claude. If the documents exceed its limit, analyze each document separately, then ask the model to compare the extracted clauses.

 
8
AI Profile
 

Ema sells software agents that carry out HR, IT and finance workflows for large companies. Its arguable position is commercial: it charges for completed tasks and business outcomes, aiming at software and services budgets rather than another per-seat assistant.

Founders:

Surojit Chatterjee and Souvik Sen founded Ema in 2023. Chatterjee was Coinbase’s chief product officer and a Google product vice president; Sen was Okta’s vice president of engineering and previously worked on machine learning at Google.

Product:

Ema coordinates agents across a customer’s existing applications to complete multi-step work such as employee support, service-desk requests and finance operations. HR, IT and finance teams pay for outcomes. As of September 23, 2026, Ema reported more than 50 active enterprise deals and over 1 million active enterprise users; published customer cases name Hitachi, Wipro and NTT DATA.

Financing:

Ema raised a $77 million Series B led by Creaegis on September 23, 2026, taking total funding to $140 million. The company did not disclose the round’s valuation, saying only that it had more than quadrupled since its 2024 round.

The risk:

Glean, UiPath, ServiceNow and Microsoft can sell agents beside systems they already control. Ema also relies on more than 150 outside models. Its reported $150 million in bookings as of September 23, 2026 covers two- and three-year contracts, and management declined to disclose annualized revenue.

Visit Ema →
 
9
The Rausschmeisser*
OpenAI will test ads beneath ChatGPT's image loading bar.

OpenAI will test visual ads beneath ChatGPT's image generation progress bar for free users in the US later this month. The placement carries an “Ad” label.

Our take: OpenAI is selling the pause while ChatGPT makes your image. Free users in the US test may get a row of scrollable product photos beneath the progress bar, an invitation to browse while their original request is still unfinished. Apparently, waiting needed a shop window.

OpenAI says ChatGPT reaches 1.2 billion people each week, and it describes advertising as a way to fund broader access to AI beyond what subscriptions alone can support. The loading bar becomes a contribution to that cause, collected in attention from people waiting for the machine to finish. Even doing nothing now comes with something to buy.

*German for the last song of the night, the one that clears the room.

FREE WEEKDAY MORNING BRIEFING

Don’t miss the next AI story that matters.
The Implicator Morning Briefing filters the AI news cycle to the stories worth your attention and explains their consequences. From San Francisco, every weekday at 4:45 a.m. Pacific, 7:45 a.m. Eastern.
Send me tomorrow’s briefing →
About five minutes. No hype. No spam.
Morning Briefing

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai