11 topics covered

Listen to today's briefing
0:00--:--

Mercor $750M Series D at $20B Valuation

What happened: AI data vendor Mercor raised a $750M Series D at a $20B valuation, as covered in Latent Space's AI News roundup.

Key details:

  • Mercor scaled to $3B run-rate revenue in 27 months, per a post by co-founder @BrendanFoody quoted in the article.
  • The company reports 5 million experts on its platform.
  • Mercor crossed $1B in revenue in February and $2B in June, according to the article.
  • Mercor says it spends 3x as much on inference as on employee salaries.

Why it matters: The article frames Mercor as a data vendor business and cites its revenue growth as evidence of AI spending flowing into the economy.

Practical takeaway: For context on AI training-data vendors, the article points to its earlier deep dive on competitor Handshake AI.

Anthropic IPO Price Range and Listing Timing

What happened: Anthropic confirmed its IPO with a per-share price range, putting its potential valuation at up to $1.4 trillion, below earlier expectations.

Key details:

  • Expected IPO price is between $70.00 and $78.00 per share, which would put the valuation as high as $1.4 trillion, according to The New York Times.
  • Anthropic has applied to list its Class A common stock on the Nasdaq Global Select Market under the symbol "ANT."
  • The Wall Street Journal reports, citing sources, that Anthropic is "aiming for its shares to start trading Thursday or Friday next week."
  • Reuters notes the IPO was previously expected to be valued at as much as $2 trillion.

Why it matters: The lower price range confirms that the company is accepting a valuation below earlier hopes amid a tech selloff and midterm-election shifts in public opinion toward AI companies and data centers.

Practical takeaway: Watch for final pricing and first-day trading around the reported Thursday-or-Friday window, as the price range may change before listing.

Adobe MAX: Agentic Creative Cloud Tools

What happened: At Adobe's MAX conference in Miami, Adobe announced agentic AI features, including third-party agent access to Creative Cloud projects and new capabilities for its Firefly AI Assistant.

Key details:

  • Local Model Context Protocol (MCP) integrations were previewed, letting external agents from providers such as OpenAI, Google, and Anthropic edit native files in Photoshop, Illustrator, Premiere, and After Effects; a private beta is set for early 2027.
  • "Always-on" background agents were previewed for recurring tasks, with no launch date given; they will come to Firefly AI Assistant first.
  • Firefly AI Assistant beta gains a Plan Mode for reviewing plans before tasks run, a Model Selector, and a Firefly Boards integration.
  • The creative agent is expanding into Premiere, including tracking masked subjects through a frame and conversational edits in the Premiere mobile app.

Why it matters: Adobe is positioning its creative tools to be driven by both its own assistant and third-party agents, though many of the features remain previews or future betas.

Practical takeaway: Creative professionals can preview the Firefly AI Assistant beta now, while the local MCP integrations for Creative Cloud are not yet available.

ChatGPT Business Agents and Business Hub

What happened: OpenAI is letting businesses represent themselves inside ChatGPT through verified "Business Agents," launched in beta alongside a set of business-focused features.

Key details:

  • Business Agents are available first to "eligible" e-commerce and online service businesses in the US.
  • Users can reach an agent when a business that created one appears in a ChatGPT response; they can ask about offerings and compare products, such as differences between two products.
  • Chats with Business Agents are clearly labeled and kept separate from the main ChatGPT conversation.
  • Companies can claim their organization and get verified through a feature called Business Hub, which also shows insights such as mentions in ChatGPT responses and top "discovery topics."
  • OpenAI isn't sharing specifics about its verification methods, which vary by business, according to VP of monetization Benji Shomair.

Why it matters: This extends ChatGPT's shopping and recommendation role by giving brands a direct, labeled presence in conversations, and OpenAI says it will explore more ways for businesses to connect with customers.

Practical takeaway: If you run an eligible US e-commerce or online service business, Business Hub is the entry point to claim your profile and set up an agent; the launch is in beta.

Big Walk and AI Disclosure Dispute at Independent Games Festival

What happened: Some Independent Games Festival (IGF) jurors said they would not vote for games that use generative AI in "any capacity," and the developer of the indie game Big Walk responded about its limited AI use.

Key details:

  • The jurors' statement says there has been "a significant increase" in games "developed with generative AI tools" among this year's entries; signatories include Sam Barlow and Lena Raine.
  • The jurors say the statement doesn't speak for IGF as an organization or for all jury members.
  • Big Walk, from House House, is a co-op adventure game; the studio says "well under 0.1%" of its code relates to LLM use, with LLM snippets copied as a starting point in some cases.
  • House House said its primary LLM use was to research third-party code issues, including a shadow-rendering fix, and that it disclosed AI usage in its IGF submission but not on its Steam page.
  • The studio said it had no AI policy when developing the game and will work on a more comprehensive policy going forward.

Why it matters: The dispute centers on how indie awards should treat minimal or edge-case AI use, which Barlow said was the "tricky edge case" that became the focus.

Practical takeaway: Developers submitting to awards may face explicit AI-use rules, so disclosure in submissions is worth checking against each festival's criteria.

Ring Drone Cam AI Indoor Drone

What happened: Ring announced the Ring Drone Cam, a flying indoor security camera, at a launch event in New York City.

Key details:

  • Will arrive January 27, 2027, for $249.99, per founder Jamie Siminoff.
  • The drone patrols the home when residents are away and reacts to triggers from Ring sensors, such as motion, a door opening, a water leak, or an alarm.
  • When triggered, it flies preprogrammed routes to a set location.
  • The Verge notes it comes almost six years after Ring announced its first flying camera, the Ring Always Home Cam.

Why it matters: The product is a consumer AI-powered home-monitoring device that depends on Ring's sensor ecosystem.

Practical takeaway: If you're a Ring user considering a drone, the stated shipping date is January 27, 2027.

eBay Blocks Third-Party AI Shopping Agents

What happened: eBay appears to have blocked third-party AI agents, including Meta's Muse, from browsing its platform.

Key details:

  • A reader told The Verge that for several weeks their Muse agent had curated a daily digest of watches for sale on eBay and similar platforms; this past weekend Muse could not pull listings and said it was blocked from the site.
  • The Verge re-created the experience with other search requests, and in some cases Muse directed users to outside sites mirroring eBay listings.
  • eBay previously updated its user agreement to ban "buy-for-me agents" and "LLM-driven bots" that place orders without human review.
  • An eBay spokesperson said: "We have restricted agent access in certain cases, including unauthorized scraping and training on our data," and said it has "ongoing discussions with potential partners." The spokesperson declined to say whether new limits were recently imposed.
  • Meta declined to comment.
  • Uber fulfilled only four rides via third-party agents in Q3, and DoorDash said "several hundred" orders came through third-party agents.

Why it matters: The report adds to a pattern of retailers restricting agents, which The Verge says raises questions about how useful shopping agents can be.

Practical takeaway: If you rely on an AI agent to track eBay listings, expect access to be unreliable; the article notes Amazon blocks external agents entirely.

Gemini Robotics 2.5 Release

What happened: Google DeepMind introduced Gemini Robotics 2.5, which it calls its most capable vision-language-action model to date, and is rolling it out to robotics partners.

Key details:

  • Evaluated across more than 200 distinct tasks; Google reports 30% to 100% higher success rate than previous baselines depending on task type, and tasks completed 20% to 30% faster.
  • In a proof of concept on Mercedes-Benz production lines with Apptronik's Apollo 2 humanoid, the robot picked and packed automotive parts with a 96% success rate.
  • Inference-time reasoning guidance boosted safe task completion around obstacles by 33% and around nearby people by 43%.
  • Google released ASIMOV-Jailbreak, an open-source benchmark of hundreds of prompts; Gemini 4 Argon achieves 99.5% jailbreak resistance on it, versus 82% for the previous baseline, Gemini Robotics ER 2.
  • Gemini 4 Argon acted as a simulation engine that helped accelerate development of the model.

Why it matters: The release targets reliability and safety for robots operating outside the lab, with the Mercedes-Benz pilot offering a concrete industrial measurement.

Practical takeaway: Robotics developers working with partners can access the model now; the ASIMOV-Jailbreak benchmark is open source for safety testing.

Gemini 4 Carbon General Availability

What happened: Google is rolling out Gemini 4 Carbon to the general public, after first releasing it to a select group of cybersecurity defenders two weeks earlier.

Key details:

  • Google calls Carbon its "strongest model" and says it "pushes the frontier even further" than Gemini 4 Argon.
  • Carbon is available today via the Gemini API in Google AI Studio and Gemini for business for all paying developers and enterprises; Google AI Ultra subscribers get it in the Gemini app, AI Mode in Search, and Antigravity.
  • Introductory pricing is $2 per million input tokens and $10 per million output tokens; the article says the costs double at the end of November.
  • Google is previewing a "swarm" feature in its Antigravity agentic development platform that turns a single prompt "into a coordinated team of agents."
  • Google says Argon will likely remain "the ideal choice for the majority of software engineering and knowledge work," with Carbon built for harder tasks.
  • It is not clear when, or if, Carbon will reach lower-tier or free users.

Why it matters: The release brings Google's top model to paying developers and positions it on price well below OpenAI's GPT-6 Astra and Anthropic's Fable 5.2, per The Verge's comparison.

Practical takeaway: Developers can switch to Carbon through the Gemini API, but should note the introductory $2/$10 pricing is scheduled to change at the end of November.

AlphaProof Code Formally Verifies Signal's Cryptography

What happened: Google DeepMind introduced AlphaProof Code, an agentic framework that formally verified the code of the Signal cryptographic protocol.

Key details:

  • Agents powered by Gemini Carbon generated over six million lines of Lean proofs; DeepMind's summary notes the proofs cover the compiled code running on devices, not just the design.
  • Only about one percent of the codebase needed a second look by human reviewers, according to the post.
  • The verification took four weeks, per DeepMind; the post says such work historically takes expert teams years.
  • To test the method, researchers reintroduced two previously fixed bugs into an isolated version of Signal's code, and the system produced machine-checked proofs that the buggy versions were insecure.
  • DeepMind says it is open-sourcing its formal verification artifacts.
  • Signal's CTO Ehren Kret said: "Using AI agents to translate live code into valid Lean theorems is a major breakthrough."

Why it matters: DeepMind calls this the first time AI has been used to formally verify a widely deployed cryptographic protocol at this depth, and the first time such a protocol's implementation has been formally verified by any method.

Practical takeaway: The verification artifacts are being open-sourced, so security researchers can inspect the proofs and the technical paper directly.

Anthropic Claude Office Suite: Meetings, Sheets, Whiteboards

What happened: Anthropic unveiled new Claude tools for non-technical knowledge workers, including Claude Meetings, Claude Sheets, and Claude Whiteboards, at an event in San Francisco.

Key details:

  • Claude Meetings can join a meeting, take notes, and lay out ideas; Claude Sheets lets Claude edit spreadsheets alongside employees; Claude Whiteboards provides a shared whiteboard; a presentation can be turned into a one-minute explainer video.
  • Whiteboards and Sheets are in beta, with Sheets available to all paid plans and Whiteboards also to all paid plans; Meetings is available for Teams and Enterprise subscriptions.
  • Claude Data (formerly Claude Dashboards, introduced in beta in October) is now generally available to all paid plans.
  • New interactive connectors let users open apps from partners such as HubSpot, Adobe, and Atlassian inside Claude, including a native Adobe PDF editor.
  • Anthropic's MCP marketplace has grown to more than 9,000 plugin options, according to the article.

Why it matters: Anthropic is extending beyond engineers, which the article notes is aimed at closing a gap for non-developer adoption, as it approaches its IPO. Competing offerings from Google and OpenAI already include meeting notes and spreadsheet features.

Practical takeaway: Paid Claude users can try Sheets and Data now; Meetings requires a Teams or Enterprise plan.