Signal SentrySearchUpdated Nov 19, 17:30 UTCPM Drop
Drop #096 · Thu, Nov 19, 2026 · Evening

Proofs for real code; agent tools move behind gateways

10-second version
  1. Anthropic says Claude proved fuzz targets in 123 open-source libraries never crash, finding long-missed bugs on the way.
  2. LangSmith and AUMOVIO show agent tool and model access moving behind one gateway with identity, policy and logs.
  3. New preprints: full ledgers stop repeated agent actions, and small-model-first cascades cut coding-agent GPU-hours up to 33%.
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 tool and model access is moving behind one governed gateway

    If you assumed
    that each agent can safely carry its own API keys and a full list of tools
    Now test
    whether any agent at your company can call a tool or model that no central policy, identity check or log covers
    This week
    List each agent's credentials and tools, and mark which ones bypass a gateway or audit log

    Based on:LangSmith Tool Gateway puts agent tool credentials, deny rules and approvals in one layerAUMOVIO routes all model access through one Databricks gateway with Entra ID sign-in2 primary sources

  2. New: machine-checked proofs, not more tests, are vouching for AI-written work

    If you assumed
    that fuzzing hours and test coverage are the strongest assurance you can get for a dependency
    Now test
    whether a formal checker exists for code or output your agents produce, and whether anything in your pipeline uses it
    This week
    Pick one critical C or C++ dependency and check whether its fuzz targets have published proofs

    Based on:Anthropic: Claude proved fuzz targets in 123 open-source libraries never crashAWS releases tools that let AI agents write and check Isabelle proofs over MCP2 primary sources

  3. New: coding-agent waste sits in hard-task routing and steps after the fix

    If you assumed
    that running a frontier model on every coding task is the safe default
    Now test
    how many of your agent's steps come after its first passing patch, and how a small model does on your easy tickets
    This week
    Route a week of easy tickets to a small model first, escalate only on failure, and compare cost and results

    Based on:Preprint: trying a small open model first cut coding-agent GPU-hours up to 33%Preprint: coding agents kept working after a correct patch, and 1–8% then undid it2 primary sources

The Drop in 60

This Drop in 50 seconds

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

Transcript

Anthropic says Claude proved code in 123 libraries won't crash. Claude patched bugs it found, then proved 239 entry points crash-free for inputs under 64 megabytes. LangSmith Tool Gateway, now in public beta, scopes an agent's tools and credentials and logs every call. A preprint found simpler agent runtimes repeated actions in 117 of 200 crash replays. Stack Overflow says Snowflake named about 14 agent practices before scaling them. Researchers say trying a small model first cut coding agent GPU hours up to 33 percent. Ask whether your key libraries have proofs, not just fuzzing hours.

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

451models priced on OpenRouter
+1,090★stars today on the top repo
4,154open roles at AI labs
9 ddays to the next AI-law deadline
Ranked by impact

Top 5 that matter

1
Security

Anthropic: Claude proved fuzz targets in 123 open-source libraries never crash

Working with Theorem, Anthropic says Claude patched bugs it found, then proved 239 fuzz entry points across 123 OSS-Fuzz libraries won't crash for any input under 64 MB, under stated RISC-V and Linux assumptions.

Why it matters: If Anthropic's results hold, proofs become a practical check on C and C++ dependencies; ask vendors and maintainers whether key fuzz targets are proven, not just fuzzed.

2
Agents

LangSmith Tool Gateway puts agent tool credentials, deny rules and approvals in one layer

In public beta, LangChain's gateway sends each MCP tool call through one URL, decides whether it runs as the user or the agent, applies deny and approval policies, and records who called what.

Why it matters: If your agents hold hand-rolled OAuth tokens or see a whole tool server, this is the pattern to copy or buy: scoped tool kits, identity per call and an audit trail.

On the Frontier watch →
3
Agents

Preprint: crashed agent tool calls repeated in 117 of 200 replays without a full runtime

A preprint injected ten fault types into an agent runtime. With a ledger, reconciliation and evidence sealing, 0 of 200 crash replays repeated an effect versus 117 for simpler setups, at about one-seventh the throughput.

Why it matters: Idempotency keys alone don't tell you whether an interrupted action happened; check that your agent runtime reconciles against the target system before it retries.

On the Frontier watch →
4
Agents

Snowflake's coding-agent rollout: name what works, then scale it, Stack Overflow reports

Stack Overflow says Snowflake SVP Vivek Raghunathan's team ran a five-stage rollout, gathered roughly 14 named practices such as one git worktree per agent, and spread them through org-wide focus weeks.

Why it matters: If adoption is high but results are uneven, measure breadth of use first and give someone senior the job of documenting what your best engineers do before redrawing roles.

5
Agents

Preprint: trying a small open model first cut coding-agent GPU-hours up to 33%

Researchers tested 36 open-weight models and say small ones fall short mainly on hard tasks. Trying a small model first and escalating only when needed cut GPU-hour costs up to 33% on SWE-bench Verified, they report.

Why it matters: Before buying a learned router, test a plain small-first fallback on your own coding tasks; the authors say it nearly matched an oracle-style router.

Agents

Agents at work

See all Agents →
Agents

AUMOVIO routes all model access through one Databricks gateway with Entra ID sign-in

Why it matters: When access ends with the corporate identity, offboarding covers AI too; check whether your agents still run on shared API keys that outlive the people who created them.

More detail: AUMOVIO routes all model access through one Databricks gateway with Entra ID sign-in

AUMOVIO says vendor-neutral model aliases let it re-point apps to new providers with no code changes, apps get access by exchanging an Entra ID token instead of API keys, and spend is charged to cost centers.

Agents

Preprint: coding agents kept working after a correct patch, and 1–8% then undid it

Why it matters: If you pay per step, test a stop check that runs the tests after each patch; extra exploration costs money and now and then reverses a correct fix.

More detail: Preprint: coding agents kept working after a correct patch, and 1–8% then undid it

Across five models, two frameworks and three SWE-bench sets, researchers report 38–82% of successful runs were right on the first source edit; for Qwen3-Coder, a median 35–37% of steps came after that.

Agents

AWS releases tools that let AI agents write and check Isabelle proofs over MCP

Why it matters: Pairing an agent with a checker that rejects bad steps as it works is a pattern teams can reuse wherever a formal check exists for the agent's output.

More detail: AWS releases tools that let AI agents write and check Isabelle proofs over MCP

Amazon Science says I/Q, I/R and Isabelle Assistant, part of its AutoCorrode suite, link an agent to a live Isabelle session; in its examples Claude Opus 4.6 produced proofs the checker accepted.

Security

Attacks and flaws

See all Security →
Security

Preprint: Kaer stops malicious agent feedback from hijacking an orchestrator's replanning

Why it matters: If your orchestrator rewrites plans from worker output, treat that output as untrusted input, the same as a web page it fetched.

More detail: Preprint: Kaer stops malicious agent feedback from hijacking an orchestrator's replanning

Researchers say an orchestrator's replanning step is open to instructions hidden in sub-agent feedback. Kaer passes failures only as fixed templates and limits plan edits; they report it beat Magentic-One and ControlValve on robustness.

Security

Canada's Cyber Centre flags fixes for TeamCity, GitHub Enterprise Server and GitLab

Why it matters: These systems hold source code and pipeline secrets; confirm self-hosted instances run the fixed releases. The advisories do not say the flaws are being exploited.

More detail: Canada's Cyber Centre flags fixes for TeamCity, GitHub Enterprise Server and GitLab

The Canadian Centre for Cyber Security listed affected versions: TeamCity before 2025.11.7 and 2026.1.3, GitHub Enterprise Server 3.19 to 3.22 before fixed point releases, and GitLab before 19.3.9, 19.4.8 and 19.5.6.

Business

Money and deals

See all Business →
BusinessFollowing since Nov 15, 2026in 9 Drops

Anthropic's IPO leaves it expected to hold about $220 billion in cash, TechCrunch says

Why it matters: For teams buying Claude, new funding shifts to public investors and bonds rather than suppliers; analysts told TechCrunch the main risk is falling behind on models.

More detail: Anthropic's IPO leaves it expected to hold about $220 billion in cash, TechCrunch says

TechCrunch says supplier-investors Amazon and Google hold stakes worth about $307.9 billion combined at $83 a share; Anthropic reported $17 billion in third-quarter revenue against $900 billion in spending commitments.

Business

IEA: data center power use could double to about 1,000 TWh a year by 2030, CNBC reports

Why it matters: Capacity plans that assume new regions arrive on schedule may slip; ask cloud and colocation providers how grid connection queues affect their delivery dates.

More detail: IEA: data center power use could double to about 1,000 TWh a year by 2030, CNBC reports

Under its Stated Policies Scenario, the IEA says grid limits, scarce equipment and local pushback could delay builds; in several regions the queue of data centers waiting to connect is larger than current peak demand.

Business

StepSecurity raises $20 million to guard dev machines, coding agents and CI pipelines

Why it matters: Supply-chain security money is following coding agents into the pipeline; if you use Harden-Runner's free tier, look at its new registry policy gate.

More detail: StepSecurity raises $20 million to guard dev machines, coding agents and CI pipelines

SecurityWeek reports the Series A, led by Elephant, brings StepSecurity's total to $23 million; its free Harden-Runner CI agent is used by over 20,000 open source projects.

Business

Study: majority-Black neighborhoods have 2.2 times the odds of hosting a data center

Why it matters: Siting is becoming a civil rights question; teams planning builds should expect calls for zoning rules and racial impact statements, both of which the study recommends.

More detail: Study: majority-Black neighborhoods have 2.2 times the odds of hosting a data center

Researchers at George Washington and American University examined 81,000 census tracts and 5,291 data centers, 404 Media reports, and found the gap held after accounting for income, land costs and transmission lines.

Laws

Law watch

See all Laws →
LawsFollowing since Nov 16, 2026in 3 Drops

Federal Register publishes Commerce rule on remote access to advanced chips

Why it matters: The official text is now public, so compliance teams can compare it with earlier press accounts of which overseas data centers need licenses.

More detail: Federal Register publishes Commerce rule on remote access to advanced chips

The Federal Register shows a rule titled Addressing Remote Access to Advanced Semiconductors from the Commerce Department's Bureau of Industry and Security, dated November 19.

Repos

Climbing fastest

See all Repos →
ReposFollowing since Nov 17, 2026in 4 Drops

rust-anything rewrites apps in Rust piece by piece, checked against originals; 4654 stars

Why it matters: It applies the same check-as-you-go idea to migrations; worth a look if a Rust port is on your roadmap, but the project is only days old.

More detail: rust-anything rewrites apps in Rust piece by piece, checked against originals; 4654 stars

MIT-licensed and created November 16, it ports CLIs, HTTP servers and libraries from six languages including TypeScript, Python and Go, checking recorded behavior at each step; up 1090 stars since our last check.

Today's Talk

What people are arguing about

See all Today's Talk →
Today's Talk

HN discusses a site saying DeepMind employees are unionising; 100 pts

Why it matters: Organized staff at a frontier lab could shape which work it takes on; watch for a statement from Google DeepMind before treating this as settled.

More detail: HN discusses a site saying DeepMind employees are unionising; 100 pts

A Hacker News thread on a site titled DeepMind Employees Are Unionising drew 100 points and 73 comments.

Today's Talk

X posts relay Trump saying the US opposes global rules that would slow AI

Why it matters: If the posts are accurate, US and EU approaches may keep diverging; plan compliance per jurisdiction rather than waiting for one shared standard.

More detail: X posts relay Trump saying the US opposes global rules that would slow AI

A Reuters post on X says Trump said Washington rejects global regulation aimed at restraining AI; Cointelegraph posted that he said it at the APEC summit, and other accounts shared clips.

Play the NewsAI ConnectionsAbout 3 minutes, built from this Drop's items.