Signal SentrySearchUpdated Nov 18, 17:30 UTCPM Drop
Drop #094 · Wed, Nov 18, 2026 · Evening

Agent code gets poisoned; banks push risk to users

10-second version
  1. Poisoned Microsoft Agent Framework releases on PyPI tried to steal cloud and CI credentials; 1.26.0 is the last clean version.
  2. Citi and PNC now put losses from customers' AI agents on the customer, TechCrunch reports, even when the tool is hacked.
  3. Field reports: Google's Sashiko reviews Linux patches at scale, and KPMG UK ran over 100 agents on an audit with humans signing off.
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. Agent losses are landing on whoever handed over the keys

    If you assumed
    that a bank or a vendor will absorb the damage when an agent holding your credentials goes wrong
    Now test
    whether any agent or agent package at your company holds a bank, cloud or pipeline credential, and who pays if it leaks
    This week
    List the agents and agent packages that touch credentials, pin their versions, and name an owner for each.

    Based on:Citi and PNC tell customers they own the losses when AI agents use their bank loginsPoisoned Microsoft Agent Framework releases on PyPI tried to steal cloud and CI keys

  2. New evidence: agent review earns trust when every finding must be proven

    If you assumed
    that one agent pass over a change or a document is enough to catch what matters
    Now test
    whether your review agents must back each finding and each dismissal with evidence, and whether conflicts survive handoffs
    This week
    Split one agent review into a finder and a separate verifier, and compare what each catches on last month's bugs.

    Based on:Google's Sashiko code-review agent: 1200 Linux reviews a day, fewer follow-up fixesPreprint: delegating to worker agents found more conflicting facts but flagged fewerKPMG UK trial: over 100 AI agents worked a FTSE 250 audit next to human auditors2 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

A poisoned agent framework, and banks change the rules. SecurityWeek reports poisoned releases of Microsoft's Agent Framework tried to steal cloud and pipeline keys. TechCrunch reports Citi and PNC now say customers own what an AI agent does with their bank login. Google says Sashiko reviews about 1200 Linux patches a day. Follow-up fixes fell to 2.2 percent. The FT reports KPMG UK put over 100 agents on an audit. A human still stepped in. Microsoft AI released seven models, plus licensed weights you can run in your own Azure. Installed version 1.26.1 or 1.26.2? Rotate your keys.

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

450models priced on OpenRouter
+1,370★stars today on the top repo
4,163open roles at AI labs
10 ddays to the next AI-law deadline
Ranked by impact

Top 5 that matter

1
Security

Poisoned Microsoft Agent Framework releases on PyPI tried to steal cloud and CI keys

SecurityWeek reports versions 1.26.1 and 1.26.2 of agent-framework, with about 190,000 monthly downloads, carried a credential stealer that runs whenever Python starts. A Microsoft maintainer confirmed their removal.

Why it matters: If any laptop, image or pipeline installed 1.26.1 or 1.26.2, rotate its cloud, Azure DevOps and GitHub Actions secrets; SecurityWeek says 1.26.0 is the last clean release.

Preliminary Way in: Two tampered releases published to PyPI; how they were published is still unknown

Controls that would have helped:

2
Agents

Citi and PNC tell customers they own the losses when AI agents use their bank logins

TechCrunch reports Citi changed its terms so customers who give third-party AI agents account access answer for what the agents do, even if the tool is hacked. PNC changed its terms too; JPMorgan and Bank of America did so last month.

Why it matters: If your product acts on a customer's bank login, the bank may not reverse its mistakes. Check whether read-only or bank-provided access can do the job instead.

3
Agents

Google's Sashiko code-review agent: 1200 Linux reviews a day, fewer follow-up fixes

Google says Sashiko, in production for seven months, splits bug-finding from verification across specialist agents. Kernel commits needing a fix within 60 days went from 3.4% to 2.2% by Linux 7.2; Google won't credit Sashiko alone.

Why it matters: Evidence that an agent reviewer can earn maintainers' trust when each finding and each dismissal must be proven in code. Try separating finder agents from verifier agents in your own review.

On the Frontier watch →
4
Agents

KPMG UK trial: over 100 AI agents worked a FTSE 250 audit next to human auditors

European Business Magazine, citing the FT, reports KPMG's head of audit said the agents sometimes probed deeper than the human team. An auditor stepped in when an agent erred, and the findings were not used to approve the accounts.

Why it matters: Even this pilot kept people accountable: KPMG says the technology is not ready for thousands of audits. Scope agent pilots with a named human who signs the result.

5
Models

Microsoft AI lists seven models, including Cyber 2 Flash, and offers licensed weights

Microsoft AI says its new Cyber 2 Flash model outscores GPT-6 Astra on CyberGym-E2E for under 2% of the per-task cost, and its Licensed Weights let customers tune a MAI model and run it in their own Azure environment.

Why it matters: These are Microsoft's own comparisons, not independent tests. If you run security or coding agents around the clock, price a MAI pilot against your current model on your own tasks.

Agents

Agents at work

See all Agents →
Agents

LangSmith ties coding-agent spend to the pull requests it produced, in public beta

Why it matters: Gives finance and engineering one unit, cost per merged PR, for judging a model switch. LangChain notes spend with no PR can still be useful review or investigation work.

More detail: LangSmith ties coding-agent spend to the pull requests it produced, in public beta

LangChain says LangSmith now ties spend in Claude Code, Codex, Cursor, Copilot and Deep Agents traces to linked GitHub or GitLab PRs, showing cost per merged PR and spend that never led to one.

Agents

Preprint: delegating to worker agents found more conflicting facts but flagged fewer

Why it matters: Splitting research across sub-agents can hide contradictions at the handoff. Test whether your lead agent sees the conflicting evidence workers found, not only their summaries.

More detail: Preprint: delegating to worker agents found more conflicting facts but flagged fewer

A preprint reports that on Harvey LAB document-review tasks across three models, forced delegation raised the share retrieving both conflicting claims from 50.2% to 68.7%, while conflict detection fell from 16.2% to 13.2%.

On the Frontier watch →
Agents

Preprint: a deployed agent system found 92 documentation gaps from 499 support tickets

Why it matters: Support tickets can feed doc fixes that also improve later agent answers, the author reports, but the scores come from LLM judges in a shadow run, not live traffic.

More detail: Preprint: a deployed agent system found 92 documentation gaps from 499 support tickets

A paper on arXiv reports TriAgent ran 43 days in shadow mode; a panel of LLM judges rated 81.1% of its checkable assertions correct. Each documentation fix went in after human review.

Security

Attacks and flaws

See all Security →
Security

Microsoft makes Perception security agents GA; Agent 365 spots 60 kinds of local agents

Why it matters: Coding agents on laptops are now treated as insiders to inventory. Check whether you can list which local agents run on your developers' machines today.

More detail: Microsoft makes Perception security agents GA; Agent 365 spots 60 kinds of local agents

SecurityWeek reports Perception's four blue-team agents are now generally available; Microsoft says its triage agent handles roughly 80% of a typical Defender alert queue. Detecting local agents like Claude Code is in preview.

Business

Money and deals

See all Business →
Business

Nvidia says its data center loan guarantees start only once customers secure financing

Why it matters: Data center plans that leaned on Nvidia backing may face tighter terms. If you rent capacity from a neocloud, ask how its next build is financed.

More detail: Nvidia says its data center loan guarantees start only once customers secure financing

TechCrunch reports interim CEO Colette Kress told analysts a guarantee, such as the $105 billion one for OpenAI's PORTS-Pike campus, lapses if a project stalls before it is built. She said Nvidia will be far more selective.

BusinessFollowing since Nov 15, 2026in 7 Drops

Anthropic weighs a bigger IPO on about $400 billion of institutional demand, CNBC says

Why it matters: Until final pricing lands, every figure here is a non-binding indication from unnamed sources. Read it as a gauge of investor appetite for AI labs, not a settled outcome.

More detail: Anthropic weighs a bigger IPO on about $400 billion of institutional demand, CNBC says

People familiar told CNBC Anthropic may price well above $80, with retail demand nearing $100 billion; the orders are not binding. Pricing is expected after Wednesday's close, with Nasdaq trading as ANT on Thursday.

Jobs

Jobs pulse

See all Jobs →
Jobs

Nvidia's interim CEO promises early pay raises and more equity to steady staff

Why it matters: Kress has called stabilizing staff a top priority. If you recruit from Nvidia, factor December raises and equity into your offers.

More detail: Nvidia's interim CEO promises early pay raises and more equity to steady staff

Wired reports Colette Kress told Nvidia's first all-hands since Huang's exit that raises and more equity arrive in December, months early; pay led the 700 questions staff submitted.

Laws

Law watch

See all Laws →
LawsFollowing since Nov 16, 2026in 2 Drops

US rule makes data centers in four Southeast Asian countries get licenses for AI chips

Why it matters: If you rent GPUs in that region, analysts quoted by CNBC expect slower, costlier builds. Check where your provider's capacity sits and who else it serves.

More detail: US rule makes data centers in four Southeast Asian countries get licenses for AI chips

CNBC reports the interim rule requires licenses for advanced AI chip buys by data centers in Singapore, Malaysia, Thailand and Indonesia; sites serving large Chinese customers have 90 days to apply. US-headquartered firms are exempt.

Repos

Climbing fastest

See all Repos →
ReposFollowing since Nov 17, 2026in 2 Drops

rust-anything ports apps to Rust step by step, checked against originals; 2967 stars

Why it matters: Checking each step against recorded behavior suits agent-driven migrations. Try it on a small CLI before trusting it with a service.

More detail: rust-anything ports apps to Rust step by step, checked against originals; 2967 stars

Per its repo description, it covers CLIs, HTTP servers and libraries in TypeScript, Python, C, Go, Ruby and Java. MIT license, created 2026-11-16; it gained 1370 stars since our last check.

Today's Talk

What people are arguing about

See all Today's Talk →
Today's Talk

HN weighs a Berlin guide site's claim it lost 75% of traffic to AI Overviews; 695 pts

Why it matters: If your product or docs rely on search traffic, track referrals from AI answers separately; this thread shows how sharply some site owners say theirs fell.

More detail: HN weighs a Berlin guide site's claim it lost 75% of traffic to AI Overviews; 695 pts

A Hacker News thread drew 695 points and 635 comments on a post by the owner of All About Berlin, who writes that Google AI Overviews took most of the site's traffic after seven years of growth.

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