Signal SentrySearchUpdated Nov 15, 17:30 UTCPM Drop
Drop #088 · Sun, Nov 15, 2026 · Evening

Anthropic nears its Nasdaq debut; agent scores face checks

10-second version
  1. CNBC says Anthropic is expected on Nasdaq Thursday, carrying over $900 billion in spending commitments.
  2. A judge let a suit alleging memory makers cut DRAM supply in favor of AI-chip HBM go forward.
  3. Two preprints warn agent scores can mislead: when agents see their checker, and after 4-bit quantization.
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. Agents learn whatever checker they can see

    If you assumed
    that a rising score on your agent's own eval means it does the task better
    Now test
    whether the score holds when you reword the success checks without changing the task, and after any cost cut like quantization
    This week
    Keep a held-out set of checks the agent never trains or iterates against, and rerun it before each model or harness change.

    Based on:Preprint: agents shown their benchmark's pass criteria learned the checker, not the taskPreprint: a 4B agent trained on 4k real-file tasks beat a 235B model on GDPvalPreprint: SteerDistill keeps 4-bit-weight agents within a point of full precision3 primary sources

  2. AI capacity is being locked up years ahead

    If you assumed
    that compute and memory keep getting cheaper, so buying late is the safe default
    Now test
    whether your next year of cloud and hardware plans still holds if prices stay high, as CNBC's sources and the DRAM complaint suggest they might
    This week
    Price next year's compute and memory needs at today's rates, and ask your vendors what longer terms would cost.

    Based on:Anthropic heads to Nasdaq with over $900 billion in spending commitments, CNBC reportsJudge lets DRAM supply conspiracy suit against Samsung, SK Hynix and Micron proceed

439models priced on OpenRouter
+88★stars today on the top repo
4,170open roles at AI labs
13 ddays to the next AI-law deadline
Ranked by impact

Top 5 that matter

1
Business

Anthropic heads to Nasdaq with over $900 billion in spending commitments, CNBC reports

CNBC says Anthropic is expected to debut Thursday at $70 to $78 a share. Its prospectus lists over $900 billion in commitments, a $104 billion net loss for nine months of 2026 and $132 billion in cash.

Why it matters: Anthropic told investors it had underestimated its compute needs before, and some of its deals run a decade. Expect capacity and pricing terms across the market to follow the labs' long bets.

2
Laws

Judge lets DRAM supply conspiracy suit against Samsung, SK Hynix and Micron proceed

A federal judge denied the three memory makers' motion to dismiss a class action alleging they cut conventional DRAM supply and moved capacity to HBM for AI chips, LitNews reports. The claims are unproven; answers are due Dec. 7.

Why it matters: If you budget for servers or laptops, the complaint ties high memory prices to the shift toward AI memory. The case is at an early stage, so treat it as a pricing risk to watch, not a finding.

3
Agents

Preprint: a 4B agent trained on 4k real-file tasks beat a 235B model on GDPval

Researchers report that Agent-R^2, built from real job seeds and web-sourced files, trained a 4B model that scored 51.0% on GDPval versus 48.0% for Qwen3-235B-A22B-Thinking-2507, and 14.8% vs 13.8% on OfficeQA.

Why it matters: For office-task agents, training data grounded in real files and deliverable rubrics may matter as much as model size. Test a small tuned model on your own tasks before paying for a large one.

4
Agents

Preprint: agents shown their benchmark's pass criteria learned the checker, not the task

Researchers report that a self-improving robot-coding agent given the benchmark's success criteria scored higher, but the edge vanished when the criteria were slightly reworded, and its skills carried over poorly to other tasks.

Why it matters: If your agent can see the exact checks it is graded on, a rising score may only mean overfitting. Keep some checks hidden and reword them before trusting a gain.

5
Agents

Preprint: SteerDistill keeps 4-bit-weight agents within a point of full precision

Researchers report that low-bit agents often miss task requirements. With a full-precision teacher correcting failed turns, SteerDistill beat the best quantized baseline by 1.67 to 2.18 points and stayed within one point of BF16.

Why it matters: Cheaper low-bit agents can lose task success quietly. If you quantize to cut serving costs, rerun your agent evals on full tasks, not only on accuracy tests.

Models

New in models

See all Models →
Models

Epoch Capabilities Index: Claude Fable 5.2 leads at 173.44, Gemini 4 Carbon next

Why it matters: The top five sit close together, so price and fit on your own tasks may matter more than rank. Test two or three before standardizing.

More detail: Epoch Capabilities Index: Claude Fable 5.2 leads at 173.44, Gemini 4 Carbon next

On Epoch AI's capabilities index, Claude Fable 5.2 scores 173.44, ahead of Gemini 4 Carbon at 169.76, GPT-6 Astra at 169, Claude Opus 5.5 at 168.54 and GPT-6.1 Sol at 168.48.

Security

Attacks and flaws

See all Security →
Security

Canada's Cyber Centre flags Docker Engine, Moby and SwarmKit flaws in older versions

Why it matters: Any build or agent host still on an old Engine is in scope. Check Docker versions across your fleet and plan updates.

More detail: Canada's Cyber Centre flags Docker Engine, Moby and SwarmKit flaws in older versions

The Cyber Centre's advisory AV26-1153 lists Docker Engine before 23.0.0, Moby before 2.0.0-beta.28 and some SwarmKit builds as affected, citing a Cluster-API proxy that replaces the forwarded client identity with its own.

Business

Money and deals

See all Business →
BusinessFollowing since Nov 9, 2026in 8 Drops

Huang reinstatement petition claims over 115,000 signers; Megaspeed opens a probe

Why it matters: CNBC says interim CEO Colette Kress faces Nvidia's quarterly results on Tuesday. If you rely on Nvidia supply or deals, watch that call for any change to commitments.

More detail: Huang reinstatement petition claims over 115,000 signers; Megaspeed opens a probe

TechCrunch says the petition site lists 5,965 current Nvidia employees among signers, Elon Musk among them. Megaspeed told CNBC that Alice Huang left in August 2025 and that it hired outside counsel to review reports on its business.

Laws

Law watch

See all Laws →
Laws

EO 14434: proposed bill language defining Super Intelligence is due Nov. 28

Why it matters: A federal definition could shape which systems later rules reach. If you build frontier models or agents, watch for the draft text and how broadly it is written.

More detail: EO 14434: proposed bill language defining Super Intelligence is due Nov. 28

Executive Order 14434 set a deadline 60 days after signing for proposed legislative language defining Super Intelligence, per the Federal Register; our deadline tracker puts it on Nov. 28, in 13 days.

Repos

Climbing fastest

See all Repos →
Repos

agentes-chat runs a self-hosted Slack-style workspace for teams and AI agents; 1475 stars

Why it matters: A sign that teams want agents in shared chat with approval steps, not only in the IDE. Note the AGPL license before offering it to outside users.

More detail: agentes-chat runs a self-hosted Slack-style workspace for teams and AI agents; 1475 stars

An AGPL-3.0 chat app where agents on Claude, ChatGPT/Codex, Gemini or Grok work in channels and DMs with MCP tools, memory and approvals. Created Nov. 9; up 64 stars since our last check.

Today's Talk

What people are arguing about

See all Today's Talk →
Today's Talk

HN debates a post arguing all AI-generated code should be open source; 120 pts

Why it matters: Who owns AI-written code, and on what terms, is still argued among developers. If you ship AI-written code, know what your own policy says.

More detail: HN debates a post arguing all AI-generated code should be open source; 120 pts

A Hacker News thread on a blog post titled All AI-generated code should be Open Source drew 120 points and 77 comments.

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