Anthropic weekly, October 12–19: a $16.8 billion quarter, a $1.5 trillion rethink, and self-replicating models
Anthropic's investor day produced a $16.8 billion quarter and a cash-flow-positive claim, but investors pushed the valuation they discuss down toward $1.5 trillion, as the company published self-replication research and pitched its own federal safety bill.
Anthropic showed investors a cash-positive quarter, watched them mark its price down toward $1.5 trillion, and published research arguing small models may soon copy themselves across consumer devices.
Funding and corporate: a bigger quarter, a smaller number
The company met prospective investors at its San Francisco headquarters on October 14 1.
It told them third-quarter revenue reached $16.8 billion, up 45% from the previous quarter, and that it was profitable on an adjusted basis 2.
Revenue more than doubled from the first quarter to the second. At 45%, growth decelerated, and some attendees said so 2.
Four people told Reuters that current and prospective investors are weighing a valuation of about $1.5 trillion 23, down from roughly $2 trillion discussed for the November sale.
Anthropic said it was cash-flow positive in the quarter, Axios reported 1. That sits against more than $500 billion of planned computing and infrastructure spending.
The meeting landed in a selloff: the S&P 500 fell more than 1% on October 14, its first decline in 53 sessions 1. CoreWeave's shares and bonds slid the next day 2.
The Financial Times reported on October 16 that Anthropic is offering Blackstone a yield above 10% on $9 billion of debt backed by five-year lease payments on the chips 45.
The same day it signed its first Canadian data-centre lease, with EverWind at Point Tupper, Nova Scotia 6.
Research and safety: a replicator that fits in fifteen TV boxes
On October 12 the Frontier Red Team published a study of autonomous replication and adaptation 7.
The demonstration split Gemma4-26B-A4B across 15 off-brand Android TV boxes bought for about $55 each. It served at 0.39 tokens per second 7.
A 9B model got inference running on 8 of the 9 hardware types tested. Only GLM-5.2, an open-weight model from Zhipu, completed the full chain, reaching all 8 test nodes in four of five trials 7.
Today's models can do the steps or fit the hardware, not both. Fitting a trend to open-weight models of 30B parameters or fewer, the authors put a model at GLM-5.2's level by the end of 2026 7.
A Kimwolf-sized botnet of more than two million devices would run about 133,000 replicas, worth at most 40 H100s 7 — slow, with no central point to switch off.
The same day brought a defensive result. New classifiers and preserved thinking cut the reasoning tokens distillers extracted by roughly 90% to 99% 8.
The false-positive rate is under 0.1%, and tens of thousands of accounts were banned, often within two hours.
The costs are stated plainly: some integrations needed updating, and blocked requests have been charged since September 24 8.
Governance: shares for the public, evaluators in the building
On October 18 Anthropic committed 5% of its shares to a new Anthropic Public Benefit Fund, subject to stockholder approval 9.
The fund gains a further 2% at the board's discretion for every 50% rise in company value above a set baseline 9.
With the co-founders' pledge and an employee match, about 20% of shares now go to the public good.
It is a wholly owned LLC rather than a charity, and its equity will not fund lobbying or political giving. Governance and valuation milestones arrive in the coming months 9.
On October 15, six METR staff received employee-like access: company hardware, office space, and internal systems close to what Anthropic's own risk teams hold 10.
METR publishes its findings and Anthropic cannot veto them, though it can require redaction in predefined categories.
SecureBio is reviewing the bioweapons chapter of the August Risk Report; talks with Hugging Face are early 10.
Customer and partners: Replit moves its default
At TechCrunch Disrupt this week, Replit chief executive Amjad Masad said Replit no longer works with Anthropic 11.
OpenAI is now the default behind Replit Agent, and Claude remains selectable. Masad cited cost, partnership and performance, and said Replit had been a top-five Anthropic customer 11.
People familiar with the relationship disputed that it had ended. Read the switch as a default change, not a severed contract 11.
Anthropic opened its fifth regional office in Singapore and signed an MOU with the Ministry of Digital Development and Information on October 14 12.
More than 1,500 GovTech Singapore engineers use Claude Code, alongside DBS, GIC, Grab and Sea. Singapore ranks second of 121 countries in Claude.ai use per head of population 12.
Legal and policy: a pause request, a rehearing, and a draft bill
New York City Comptroller Mark Levine wrote to Dario Amodei on October 14, urging Anthropic to pause the offering until it addresses safety and governance concerns 13.
Levine advises city pension funds holding more than $325 billion, which have $7 million of exposure to Anthropic through private equity. He asked for a written reply 14.
Anthropic asked the US Court of Appeals for the DC Circuit on October 13 to rehear the case upholding the Pentagon's supply-chain-risk designation 15.
Its argument: a statute aimed at hostile foreign actors cannot presumptively reach every contractor. A September 2-1 ruling held that Anthropic's safeguards could compromise military operations 15.
Politico reported on October 16 that Anthropic has quietly pitched its own draft federal safety bill 16.
Under it, government could block a model only for a foreseeable, imminent risk of over $1 billion in damage, or of killing or seriously injuring more than 50 people. Evaluators registered with the Commerce Department would do the assessing, reporting imminent risk within 72 hours.
Only Commerce could seek a judge's order to suspend a model; Anthropic says the draft reflects its Advanced AI Framework 16.
For the midterms, Anthropic measured 250,000 US conversations from September 21 to 27: 5.75% touched politics, 0.51% elections, and 0.01% asked how to vote 17.
Product and platform: small releases, one revealing report
OpenTelemetry export arrived in beta on October 14, Managed Agents gained scheduled deployments on Claude Platform on AWS, and fast mode left research preview for Opus 5.5 18.
Anthropic's own workforce report found junior technical hires falling from 27% to 9% of technical hiring since early 2025, and meeting time per employee up 20% 1920.
What did not happen, and what to watch
The week brought no new model, no equity round, no leadership change and no court ruling naming Anthropic. The Federal Trade Commission's industry-wide investigation produced no new public step.
- November: the listing remains reported for the month, with marketing possibly from the week of November 9 — at a price investors now put near $1.5 trillion rather than $2 trillion 3.
- November 12: the updated Usage Policy takes effect 21.
- November 30: Claude Sonnet 4.5 retires on the Claude API 18.
- Undated: METR's first published assessments, the Public Benefit Fund's governance and valuation milestones, and the comptroller's requested reply.
References
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9Anthropic: Investing in the public benefit
anthropic.com
- 10Anthropic: An update on independent evaluators
anthropic.com
- 11
- 12
- 13
- 14
- 15
- 16
- 17Anthropic: Preparing for the 2026 US midterms
anthropic.com
- 18Claude Platform release notes
docs.claude.com
- 19
- 20Anthropic: How employment is changing at Anthropic
anthropic.com
- 21Anthropic: 2026 Usage Policy update
anthropic.com
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.