Drop #013 · Fri, Oct 9 · Morning

Anthropic rewrites its rules; OpenAI's revenue resets

10-second version

Anthropic's Usage Policy update, effective November 12, sharpens weapons and surveillance rules and bans sustained abuse of Claude.
OpenAI told investors it hit about $50B in annualized revenue, below the $68B widely reported; AI stocks fell, CNBC says.
Anthropic offers free OSS Scanner security scans; Goodfire launches cheap probes that watch agents from the inside.

443models priced on OpenRouter

+128★stars today on the top repo

4,133open roles at AI labs

50 ddays to the next AI-law deadline

Ranked by impact

Top 5 that matter

Agents

Anthropic updates its Usage Policy and bans sustained abuse of Claude

Anthropic says the update, effective November 12, adds a section on deceptive campaigns, sharpens weapons and surveillance rules, requires that Claude-driven hardware can be stopped, and bans needless cruelty to its models.

Why it matters: Builders using Claude in health, finance or legal decisions still need a qualified human in the loop who can change its recommendations, and Anthropic now spells out which uses are covered.

Anthropic · Oct 8 The Verge · Oct 8

Jobs

OpenAI told investors its annualized revenue is about $50B, not $68B, CNBC says

CNBC confirmed an FT report that OpenAI hit roughly $50 billion in annualized revenue at the end of September. Nvidia fell 3%, Oracle nearly 6% and CoreWeave nearly 8% on Thursday.

Why it matters: A person familiar told CNBC the $68 billion figure included partners' gross revenue. Lab revenue comparisons depend on how each company counts, TechCrunch notes.

CNBC · Oct 8 TechCrunch · Oct 8 Hacker News · Oct 8

Agents

Anthropic launches a Cyber Mission with free OSS Scanner security scans

Anthropic says opt-in open-source projects get free periodic scans from its strongest models, sent without human review. A new program brings Claude to infrastructure security firms including CrowdStrike and Dragos.

Why it matters: Anthropic expects a true-positive rate above 90%, so some reports will be wrong. Maintainers still need to triage, and The Verge notes projects already struggle with AI bug reports.

Anthropic · Oct 8 The Verge · Oct 8

Agents

Google's Gemini agent gets its own Workspace account and audit trail, TechCrunch says

Google says the agent takes objectives, plans work, connects to tools like Slack, Jira and Snowflake or any MCP server, and lets users pick Anthropic's Claude models. It starts with businesses.

Why it matters: An agent with its own email address, writing an audit trail attributed to itself rather than a person, changes how teams grant access and trace who did what.

TechCrunch · Oct 8 The Verge · Oct 8

Agents

Goodfire launches probes that watch agents from inside the model, on Baseten

TechCrunch says Goodfire's probes read a model's internal signals and caught 93% of malicious hacking sessions in tests on Kimi K3. Baseten customers choose to log, send for human review or refuse.

Why it matters: Goodfire says monitoring about 1 million exchanges cost roughly $185, versus about $200,000 for a top-tier model checking every step, making long agent runs cheaper to watch.

TechCrunch · Oct 8

Models

New in models

See all Models →

Models

Claude Opus 5.5 tops LMArena's WebDev board as of October 6

On LMArena's webdev board as of 2026-10-06, claude-opus-5.5-max scores 1813.6, ahead of gpt-6-astra-max at 1787.596 and claude-sonnet-5.5-xhigh at 1772.752.

Why it matters: Crowd ratings on web-building tasks are a check on vendor benchmarks when choosing a coding model.

LMArena (CC BY 4.0) · Oct 6

Agents

Agents at work

See all Agents →

Agents

AWS shows agents paying per inference under spending caps set outside the model

AWS says Incarna used AgentCore payments to let agents pay BlockRun per call over x402 in USDC. Agents processed over 1,000 payments from $0.001 to $0.05 per call during the beta.

Why it matters: AWS says session limits are enforced by the infrastructure, so even a manipulated prompt can't push an agent past the cap its customer set.

AWS Machine Learning Blog · Oct 8

Jobs

Jobs pulse

See all Jobs →

Jobs

Anthropic commits $150 million over three years to the Genesis Mission

Anthropic says the money makes Claude available to more than 15 federal agencies, including NASA, NIH and NSF, with Claude and API credits for several hundred research projects.

Why it matters: Federal science agencies are becoming a sizable customer base for frontier labs, with training and support bundled in.

Anthropic · Oct 8

Jobs

Arena raises $200M at a $3.1B valuation and adds an alignment leaderboard

TechCrunch says Lightspeed and Khosla led the round. Arena said it reached $100 million in annualized run-rate revenue in June, up from $30 million at its January Series A.

Why it matters: Arena's new alignment board ranks models on unauthorized actions and claiming tasks they didn't do, useful signals for anyone picking an agent model.

TechCrunch · Oct 8

Jobs

Nvidia-backed Firmus withdraws its IPO, citing market volatility, CNBC says

The Australian AI data center operator had reportedly sought $5 billion at about a $30.6 billion valuation. Firmus says it will now pursue capital from private markets.

Why it matters: CNBC says it would have been the second-largest new share sale in Australia's history. Firmus said the terms didn't reflect its business, a signal for other compute builders raising money.

CNBC · Oct 9

Jobs

14 AI companies list 4,133 open roles, led by Databricks and OpenAI

Databricks lists 893 open roles and OpenAI 817, with Anthropic at 646. OpenAI's biggest category is Go To Market with 146; Anthropic's is Sales with 120.

Why it matters: Sales and go-to-market hiring at the biggest labs shows where they are spending: selling AI to businesses.

OpenAI jobs · Oct 9 Anthropic jobs · Oct 9

Laws

Law watch

See all Laws →

Laws

USA Today Co. sues OpenAI, seeking more than $250 million, The Verge says

The publisher and several local papers it owns allege OpenAI copied hundreds of thousands of articles to train its models. The Verge lists earlier suits from The New York Times, Ziff Davis and Britannica.

Why it matters: Each publisher suit adds to the legal and licensing pressure around training data. OpenAI didn't immediately respond to The Verge.

The Verge · Oct 8 Forbes on Bluesky · Oct 8

Laws

US suspends Microsoft and Adobe from its green card labor program, CNBC says

Labor Secretary Keith Sonderling said the department won't process new applications from them or from Cognizant, Infosys, Capgemini, Tata, Wipro and HCL. Microsoft says most of its US staff are Americans.

Why it matters: Software employers that sponsor foreign hires face a tougher federal stance on permanent residency applications.

CNBC · Oct 8

Laws

Executive Order 14434's 'Super Intelligence' definition deadline is 50 days out

Our laws registry lists November 28 as the deadline, 60 days after signing, for proposed legislative language defining 'Super Intelligence' under the order.

Why it matters: How the term is defined may shape which AI systems future federal rules cover.

Federal Register · Oct 2

Repos

Climbing fastest

See all Repos →

Repos

jiwoochris/artex-ko adds 128 stars since our last check

A Korean localization of the ARTEX AI autonomous penetration-testing framework, licensed AGPL-3.0. It has 713 stars and was created on 2026-10-03.

Why it matters: Autonomous pentest tools are spreading in local-language versions; they belong only on systems you are authorized to test.

GitHub · Oct 9

Repos

strands-agents/box adds 55 stars since our last check

Box runs AI agents in a sandbox that restricts what they can execute, read, write and reach on the network, with default-deny policies and credential injection. Apache-2.0, 239 stars.

Why it matters: Keeping secrets outside the agent and denying by default limits the damage if an agent is tricked.

GitHub · Oct 9

Today's Talk

What people are arguing about

See all Today's Talk →

Today's Talk

Fired OpenAI safety researchers dispute misconduct claims in an open letter

TechCrunch says the three researchers deny mishandling information and warn of a chilling effect. OpenAI cites a pattern of misconduct and shared a memo denying retaliation. Ro Khanna posted on X about whistleblower protections.

Why it matters: How labs treat staff who work with outside safety evaluators shapes how much independent scrutiny frontier models get.

TechCrunch · Oct 8 @RoKhanna on X · Oct 8

Today's Talk

Hacker News debates why DeepSeek 4.1 Flash isn't worrying frontier labs

A blog post drew 457 points and 396 comments on Hacker News. Its author says they used DeepSeek 4.1 Flash heavily for a month and often couldn't tell it apart from Opus.

Why it matters: The author describes routing most work to a cheap model and calling in Opus 5.5 for final code review, a cost pattern many teams are weighing.

Hacker News · Oct 8 dgt.is blog · Oct 7

Today's Talk

TechCrunch: OpenAI's math release falls short of mathematicians' standards

TechCrunch says only 10 of 719 manuscripts included the model's chain of thought, and a new paper found discrepancies in the Lean code behind a Navier-Stokes proof. A related HN thread drew 108 comments.

Why it matters: The paper's authors say self-formalized AI proofs shouldn't be trusted without peer review, a caution for any AI output checked only by AI.

TechCrunch · Oct 8 Hacker News · Oct 8

Today's Talk

Anthropic's ban on abusing Claude splits reactions on Bluesky and HN

A Hacker News thread drew 148 comments. On Bluesky, @kattenbarge.bsky.social posted that tech firms will moderate violence against AI first, while @hailey.at posted in defense of Anthropic's record.

Why it matters: Anthropic says the rule covers only extreme, repeated cruelty, not frustration, dark creative themes or model testing, so most users and red teams are unaffected.

Hacker News · Oct 8 Bluesky · Oct 8 Anthropic · Oct 8

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