Signal SentrySearchUpdated Nov 12, 17:30 UTCPM Drop
Drop #082 · Thu, Nov 12, 2026 · Evening

Huang quits Nvidia; stolen tokens hit GitHub repos

10-second version
  1. Jensen Huang resigned as Nvidia CEO; CFO Colette Kress is interim chief and Harvey Jones becomes the first board chair.
  2. Stolen developer and bot tokens let an attacker plant credential stealers in GitHub repos; rotate CI, cloud and AI keys.
  3. Databricks and AWS show agents acting as the asking user; a preprint finds no finance agent session passed its checks.
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. Valid stolen tokens, not software bugs, are getting attackers into pipelines

    If you assumed
    that keeping CI software patched keeps attackers out of your build pipeline
    Now test
    whether a stolen developer or bot token could edit workflows or install hooks in your repos without anyone reviewing it
    This week
    List every token that can push to your repos or CI, then shorten its life and narrow what it can touch

    Based on:Stolen tokens let an attacker plant stealers in GitHub repos, security firms sayFireworks AI was breached twice in October, exposing some Notion AI chats

  2. Who an agent acts for is being fixed in code, outside the model

    If you assumed
    that an agent's tools already know which user or tenant they are serving
    Now test
    whether any agent tool lets the model fill in the user, tenant or identity it acts as; new this run, vendors now ship patterns that set it from the login
    This week
    Find tool arguments that name a user or tenant and fill them from the verified login, not the model

    Based on:AWS: don't let the model decide which user an agent searches asDatabricks makes Genie in Slack generally available, acting with each asker's permissionsAWS shows A2A agents working across two companies' accounts with no shared secrets3 primary sources

  3. Agent work that looks finished can still fail a hard check

    If you assumed
    that a user signing off on an agent's output means the work is correct
    Now test
    whether your agent evaluations recompute results and compare the agent with a simple rule-based baseline over long runs
    This week
    Add one program check and one rule-based baseline to the next agent evaluation your team runs

    Based on:Preprint: no finance agent session passed FINCOSMOS's checks across 150 triesPreprint: four of five LLM agents ran 20 vending machines worse than a simple rule2 primary sources

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

Nvidia's CEO is out. Jensen Huang resigned as Nvidia's CEO, effective immediately. CFO Colette Kress is now interim chief. Security firms say an attacker used stolen developer and bot tokens to plant credential stealers in GitHub repositories. No GitHub Actions bug was needed. Apache patched a critical HTTP/2 flaw first reported by Anthropic's scanner. Update to 2.4.71. Databricks made Genie in Slack generally available. It acts with each asker's permissions. A preprint found no finance agent session passed its checks across 150 tries. Check which tokens can edit your build workflows, and make them short-lived.

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

440models priced on OpenRouter
+546★stars today on the top repo
4,167open roles at AI labs
16 ddays to the next AI-law deadline
Ranked by impact

Top 5 that matter

1
BusinessFollowing since Nov 9, 2026in 5 Drops

Jensen Huang resigns as Nvidia CEO; CFO Colette Kress named interim chief

Nvidia's board says Huang quit as CEO and director, effective at once. Ex-Synopsys chief Harvey Jones becomes its first board chair, CNBC reports, and the search for a permanent CEO covers inside and outside candidates.

Why it matters: If your GPU supply or roadmap rests on Nvidia, watch what Kress tells staff at the November 18 company meeting, per The Verge, before changing plans.

2
Security

Stolen tokens let an attacker plant stealers in GitHub repos, security firms say

Aikido counts about 4,800 compromised accounts probing 61,764 repositories since November 10, SecurityWeek reports; the stealer ran in over 1,800. StepSecurity says no GitHub Actions bug was used, only valid developer and bot tokens.

Why it matters: Where the stealer ran, Aikido says to treat every reachable secret as stolen. Check for hidden workflow edits and install hooks, and rotate CI, cloud and AI keys.

Preliminary Way in: valid stolen developer and bot tokens, then CI workflows, build scripts and npm install hooks

Controls that would have helped:

3
Security

Apache patches a critical HTTP/2 flaw that Anthropic's scanner and Google found

SecurityWeek reports HTTP Server 2.4.71 fixes CVE-2026-119618, a mod_http2 race Red Hat rates 9.8 and says allows unauthenticated code execution. Apache credits Anthropic's OSS Scanner first, then Google researchers.

Why it matters: Patch servers running 2.4.55 through 2.4.70 with HTTP/2 enabled, or switch HTTP/2 off until you can; Wiz saw no in-the-wild exploitation when it published.

4
Agents

Databricks makes Genie in Slack generally available, acting with each asker's permissions

Databricks says Genie answers in Slack DMs and channels under the asking user's identity and Unity Catalog rights, and can edit documents, tickets and records via MCP connectors only after that user approves each change.

Why it matters: A ready pattern for chat agents: actions run as the person asking, not a shared bot. Decide whether channel answers default to private and which MCP tools need approval.

5
Agents

Preprint: no finance agent session passed FINCOSMOS's checks across 150 tries

Researchers report testing four coding agents and a tool-free model in 150 sessions; none was accepted, and every run the simulated user approved still failed a program check. Agents raised only 54-64% of unstated details.

Why it matters: A finished-looking report can hide future data or unchecked numbers. Before trusting an agent's financial analysis, recompute its metrics and confirm the data was available at the time.

Agents

Agents at work

See all Agents →
Agents

AWS shows A2A agents working across two companies' accounts with no shared secrets

Why it matters: Vendors exposing agents to customers get a documented path using each side's own identity provider. Check what your Agent Card reveals about internal tools.

More detail: AWS shows A2A agents working across two companies' accounts with no shared secrets

AWS's walkthrough links a LangChain orchestrator and a Strands SaaS agent in separate accounts over A2A, with OAuth both ways and only public values swapped at onboarding. The SaaS agent publishes one curated skill.

On the Frontier watch →
Agents

Preprint: four of five LLM agents ran 20 vending machines worse than a simple rule

Why it matters: Before handing an agent ongoing operations, test it against a plain rule-based baseline over a long run, and count inference cost as well as returns.

More detail: Preprint: four of five LLM agents ran 20 vending machines worse than a simple rule

Researchers report AutoVend, where one agent runs 20 machines across five site types for 180 simulated days. Four of five agents trailed a rule-based inventory baseline, and return per inference dollar ranked models differently.

Security

Attacks and flaws

See all Security →
SecurityFollowing since Nov 10, 2026in 2 Drops

Fireworks AI was breached twice in October, exposing some Notion AI chats

Why it matters: Notion tells customers to treat the chats as exposed and rotate any credentials pasted into Notion AI. Ask your AI vendors which inference providers see your prompts.

More detail: Fireworks AI was breached twice in October, exposing some Notion AI chats

SecurityWeek reports one intrusion used compromised internal credentials and another, on October 24, ran an unauthorized workload via a cloud provider hosting Fireworks R&D machines. Notion says under 1% of workspaces were hit.

Preliminary Way in: compromised internal credentials, then a third-party cloud provider hosting R&D machines

Controls that would have helped:

Security

Critical Red Hat Quay flaw lets attackers forge tokens for any repository

Why it matters: If you run Quay, restrict access to its key service and look for unknown signing keys: Red Hat says password changes and new robot tokens won't revoke a planted key.

More detail: Critical Red Hat Quay flaw lets attackers forge tokens for any repository

SecurityWeek reports CVE-2026-123710, rated 9.8, lets an attacker plant a signing key in Quay's key service and mint read-write tokens for any repository. Fixes were merged on GitHub November 11; no fixed version is named yet.

Business

Money and deals

See all Business →
Business

Nebius lifts 2026 capex to $27 billion and plans $40 billion to $45 billion for 2027

Why it matters: If you rent GPUs from neoclouds, weigh each provider's funding and build timeline along with price: Nebius plans to spend far more than this year's revenue.

More detail: Nebius lifts 2026 capex to $27 billion and plans $40 billion to $45 billion for 2027

CNBC reports Q3 revenue of $980.8 million, up 571%, with 2026 revenue at the low end of $3 billion to $3.4 billion. Nebius guides 2027 revenue of $12 billion to $13 billion and raised $5.75 billion in debt in the quarter.

Jobs

Jobs pulse

See all Jobs →
Jobs

AI squeezes consulting's billable hours and junior work, firms tell CNBC

Why it matters: If you hire or train juniors, plan how they build judgment when AI does entry-level research, analysis and slides, and check whether your vendors still bill by the hour.

More detail: AI squeezes consulting's billable hours and junior work, firms tell CNBC

KPMG's consulting chief tells CNBC paid advice has all but vanished and outcome pricing is hard to set. EY says AI now drafts first-pass audit checks once done by juniors, and it is adding data and product roles.

Laws

Law watch

See all Laws →
LawsFollowing since the Morning Dropin 2 Drops

Huang scandal fuels a push to put chip curbs in the defense bill, CNBC reports

Why it matters: If you rent GPU capacity outside the US to Chinese customers, map that exposure now: Eurasia Group's Meale tells CNBC compliance risk would rise if such bills pass.

More detail: Huang scandal fuels a push to put chip curbs in the defense bill, CNBC reports

CNBC reports lawmakers want the AI Overwatch, Chip Security and MATCH Acts in this year's NDAA, and cites The Information that Commerce is finalizing a rule to stop Chinese firms remotely using advanced chips abroad.

Repos

Climbing fastest

See all Repos →
Repos

agentes-chat runs a self-hosted Slack alternative for teams and AI agents; 942 stars

Why it matters: Another sign teams want agents inside shared chat with approvals; review the AGPL-3.0 terms before building on it.

More detail: agentes-chat runs a self-hosted Slack alternative for teams and AI agents; 942 stars

An AGPL-3.0 project created November 9 where agents on Claude, ChatGPT/Codex, Gemini or Grok work in channels and DMs with MCP tools, memory and approvals. It gained 546 stars since our last check.

Today's Talk

What people are arguing about

See all Today's Talk →
Today's Talk

Bluesky posters say airlines were never going to let AI agents book flights

Why it matters: The reaction sets expectations for consumer agents: if your agent books or buys for users, check each merchant's terms before promising it.

More detail: Bluesky posters say airlines were never going to let AI agents book flights

After United barred third-party agents, a post by @justinhendrix.bsky.social mocking agent booking pitches drew 490 likes, and @tressiemcphd.bsky.social argued airlines' pricing power made such bans predictable.

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