Signal SentrySearchUpdated Nov 11, 05:30 UTCAM Drop
Drop #079 · Wed, Nov 11, 2026 · Morning

Gemini 4 Carbon goes wide; Grok Build links leak data

10-second version
  1. Google opened Gemini 4 Carbon to paying developers at $2/$10 per million tokens and previewed agent swarms in Antigravity.
  2. Claroc says Grok Build share links leaked company data through public DNS; Microsoft shows a backdoor via a few edited neurons.
  3. New York opened RAISE Act registration for big frontier labs, and Australia plans mandatory AI incident reporting, per Reuters.
Our read

What changed

Every signal →

Our reading of this Drop's sources: what to rethink, and one thing worth checking this week. Not a prediction.

  1. Sites are choosing which agents get in, and on what terms

    If you assumed
    that an agent can reach any public site its user can
    Now test
    whether each site your agents rely on blocks outside agents, or offers its own verified agent instead
    This week
    List the outside sites your agents browse and note which ones blocked them this month.

    Based on:eBay appears to block third-party AI agents such as Meta's Muse, The Verge reportsOpenAI launches Businesses in ChatGPT beta, with verified Business AgentsLinux Foundation takes on Agent Name Service, a DNS-based identity standard for agents2 primary sources

  2. AI app builders can publish internal tools to the open web

    If you assumed
    that an app built with AI stays behind company sign-in like the tool it replaces
    Now test
    whether apps your staff built in AI app builders open for anyone holding the link, a step past the retention defaults we flagged
    This week
    Ask staff which AI-built apps they have shared, and switch each one to invited users only.

    Based on:Grok Build share links exposed thousands of apps' data via public DNS, Claroc says

The Drop in 60

This Drop in 52 seconds

Tap to play. It never plays by itself, and the captions are in the picture.

Transcript

A new frontier model, and a leak in plain sight. Google made Gemini 4 Carbon broadly available, at $2 per million input tokens for now. Claroc says Grok Build share links showed up in public DNS records. It found 43,601 of them. New York opened RAISE Act registration. Big frontier labs must sign up by January 1, 2027. Microsoft researchers say editing fewer than 10 dormant neurons can backdoor an agent's model. TCS says its testing agents cut effort 88 percent, in its own benchmark. Check the sharing defaults on every AI app builder your team uses.

Made with AI from this Drop's sourced items. The facts come from the sources linked on each item. Script: Claude · Voice: OpenAI gpt-4o-mini-tts · Visuals: motion graphics

441models priced on OpenRouter
+75★stars today on the top repo
4,184open roles at AI labs
17 ddays to the next AI-law deadline
Ranked by impact

Top 5 that matter

1
Models

Google opens Gemini 4 Carbon to paying developers at $2/$10 per million tokens

Google says Carbon, its strongest model, is now in the Gemini API and Gemini for business, and it is previewing /swarm in Antigravity to turn one prompt into a team of agents. The Verge says prices double after the intro period.

Why it matters: Price agent workloads on Carbon now, but budget for the higher rate; The Verge notes a low per-token price doesn't fix task cost, since models burn different token counts.

3
Laws

New York opens RAISE Act registration; big frontier labs must sign up by Jan. 1, 2027

TechCrunch reports New York opened a portal for large frontier developers, those with over $500 million in annual revenue. From the new year, the law requires published safety frameworks and critical incident reports within 72 hours.

Why it matters: If you build on Anthropic or OpenAI, expect more public safety frameworks and incident reports from them; TechCrunch says fines can reach $1 million for a first violation.

4
Security

Microsoft: editing under 10 dormant neurons can backdoor an AI agent without retraining

Microsoft researchers describe Lazarus, which rewires rarely firing neurons so a model passes evals until a trigger, such as a date, makes the agent run attacker code. They say spotting it needs a diff against the original weights.

Why it matters: Treat model weights like code you deploy: check where each checkpoint came from and that it is unchanged, and limit what agents with terminal access can run.

Preliminary Way in: Write access to a model checkpoint via a compromised account, pipeline or storage

Controls that would have helped:

5
Agents

TCS says its agentic test orchestrator cut test effort 88% on about 100 scenarios

In a controlled internal benchmark, TCS says about 100 BFSI test scenarios took roughly 24 hours with its AgentCore-based agents versus about 200 hours by hand. A deterministic supervisor routes work, and fixes get bounded retries.

Why it matters: A pattern worth copying: split testing into narrow agents, cap retries and reject oversized fixes. The numbers come from TCS's own benchmark on one workload.

Models

New in models

See all Models →
Models

Kimi K3.1 Pro Highspeed lands on Amazon Bedrock with zero data retention

Why it matters: Teams wanting an open-weight model for agent loops can try it by swapping a model ID; check whether Flex latency suits background jobs.

More detail: Kimi K3.1 Pro Highspeed lands on Amazon Bedrock with zero data retention

AWS says Moonshot AI's K3.1 Pro Highspeed keeps a 1-million-token context and, per Moonshot, runs 2–3 times faster than K3.1 Pro. Data stays inside AWS and isn't shared with Moonshot; a Flex tier costs half the Standard price.

Agents

Agents at work

See all Agents →
Agents

Linux Foundation takes on Agent Name Service, a DNS-based identity standard for agents

Why it matters: Before letting outside agents call your systems, decide how you'd verify who runs them; ANS is one open option to watch, with early support in AGNTCY's directory.

More detail: Linux Foundation takes on Agent Name Service, a DNS-based identity standard for agents

The Linux Foundation says ANS builds agent naming, registration and verification on DNS and web certificates, in a seven-document spec. It lists 11 backers, including Cisco, Cloudflare, GoDaddy and Salesforce.

On the Frontier watch →
Agents

NVIDIA donates OpenShell, its agent security runtime, to the CNCF

Why it matters: Teams writing their own agent policy layer can now weigh a community-governed one before building more in-house.

More detail: NVIDIA donates OpenShell, its agent security runtime, to the CNCF

NVIDIA says OpenShell applies fixed policies to agents' tool use, inference calls and workloads, and hundreds of organizations use it. It is now a CNCF Sandbox project; NVIDIA and Google are linking it to Agent Substrate.

Agents

Anthropic puts Claude into video calls with Claude Meetings, plus Sheets and Whiteboards

Why it matters: Decide who may invite an AI into meetings and which connected apps it can query during a call before the beta reaches your staff.

More detail: Anthropic puts Claude into video calls with Claude Meetings, plus Sheets and Whiteboards

TechCrunch says Claude Meetings, in beta on Team and Enterprise with Google Meet, joins calls as a visible participant and works on tasks live. Admins must approve before it enters a call with outside guests, and it exits if one arrives.

Agents

eBay appears to block third-party AI agents such as Meta's Muse, The Verge reports

Why it matters: If your product relies on an agent browsing retail sites, test it often against each one; per The Verge, Amazon blocks outside agents while Shopify opened its catalog.

More detail: eBay appears to block third-party AI agents such as Meta's Muse, The Verge reports

A reader's Muse agent stopped pulling eBay listings, and The Verge reproduced the block. eBay says it restricts agents in some cases. The Verge cites Uber's four Q3 rides via third-party agents as a sign of thin use.

Agents

OpenAI launches Businesses in ChatGPT beta, with verified Business Agents

Why it matters: Brands may need an official agent inside ChatGPT, not just a website; settle who on your team would own its instructions and answers.

More detail: OpenAI launches Businesses in ChatGPT beta, with verified Business Agents

The Verge says eligible US e-commerce and online service firms can get verified in Business Hub and set up an agent that answers shoppers in labeled ChatGPT chats. OpenAI isn't detailing its verification methods.

Business

Money and deals

See all Business →
Business

CoreWeave raises its 2026 revenue outlook as its AI backlog reaches $132.5 billion

Why it matters: CoreWeave's CEO says GPU prices may stay high if data center building slows; factor that into capacity contracts, and watch how neoclouds pay for growth.

More detail: CoreWeave raises its 2026 revenue outlook as its AI backlog reaches $132.5 billion

CNBC reports Q3 revenue of $3.66 billion, up 168%, a 2026 forecast of $13 billion to $13.3 billion, and a backlog up from $104 billion in June. At current yields it won't issue more parent-level unsecured debt through 2027.

BusinessFollowing since Nov 10, 2026in 3 Drops

Anthropic opens an Australian retail offer as it seeks up to $64.4 billion in its IPO

Why it matters: A listed Anthropic will file public results that vendor-risk teams relying on Claude can read; TechCrunch says the founders keep 51.1% of the votes.

More detail: Anthropic opens an Australian retail offer as it seeks up to $64.4 billion in its IPO

CNBC says the Australian retail offer opened Wednesday and closes Nov. 17, ahead of a Nasdaq listing as ANT, alongside a $26 billion private placement. TechCrunch says each of the seven founders plans to sell 3,378,378 shares.

Laws

Law watch

See all Laws →
Laws

Australia plans an AI safety regulator and mandatory AI incident reporting, Reuters says

Why it matters: Teams running agents in Australia should know how they'd detect and report an agent incident; CNBC says OpenAI and Anthropic told parliament they'd back mandatory breach reporting.

More detail: Australia plans an AI safety regulator and mandatory AI incident reporting, Reuters says

CNBC, citing Reuters, says PM Anthony Albanese plans required reporting of some AI incidents and a regulator that can impose penalties, after an OpenAI agent got unauthorized access to a Medicare stats site.

Repos

Climbing fastest

See all Repos →
Repos

lot-map shows Claude Code's tool calls and subagents live beside the chat; 108 stars

Why it matters: Useful if you want to see what a coding agent actually did, rather than relying on its own summary.

More detail: lot-map shows Claude Code's tool calls and subagents live beside the chat; 108 stars

A local-only map that lights up Claude Code's tool calls, model steps, hooks, skills and subagents as they happen, in five languages. MIT license, created 2026-11-09; about 79 stars a day by our count.

Today's Talk

What people are arguing about

See all Today's Talk →
Today's Talk

HN debates VulnCheck's tally: 217 fixes from 37,966 Anthropic Glasswing findings

Why it matters: If you maintain open source, plan triage time before opting into AI scanners; VulnCheck argues the validation burden has shifted onto maintainers.

More detail: HN debates VulnCheck's tally: 217 fixes from 37,966 Anthropic Glasswing findings

VulnCheck says it tracks 442 CVEs tied to Anthropic, 2 exploited in the wild, and that curl marked 19 of 34 published scanner reports not a vulnerability. The HN thread drew 327 points and 147 comments.

Play the NewsShip It or Block ItAbout 2 minutes, built from this Drop's items.