Automattic board sidelines Mullenweg; AI agents break RSA-260 for $400K

Marcus here.

Three stories today, and each one turns on control sitting somewhere other than where it was written down.

Automattic's board put Matt Mullenweg on paid leave with 50 minutes' notice and handed the CEO job to its finance chief. He still runs WordPress.org, which the board does not own.

Andrew Tulloch waited for Meta to ship Muse, then left, eleven months after Zuckerberg chased him with an offer reported at $1.5 billion.

And Cognition's agents rewrote a decades-old sieve to factor a 260-digit number for $400,000. The same math puts RSA-1024 at about $30 million.

Stay curious,

Marcus Schuler

Automattic's board voted on September 9 to place co-founder Matt Mullenweg on paid leave and named finance chief Mark Davies interim CEO.

Mullenweg said he received the board resolution 50 minutes before the directors met, asked for time to consult independent legal counsel, and was refused. He voted against it and remains a director.

Toni Schneider, who ran Automattic from 2006 to 2014, confirmed the decision on the company's internal message board. Mullenweg named him among the directors who voted for the leave, alongside Davies, Ann Dunwoody and Sue Decker.

The change lands during Automattic's continuing lawsuit with hosting rival WP Engine. Mullenweg keeps personal control of WordPress.org, the distribution servers that deliver plugin and theme updates, which Automattic does not own.

Why This Matters:

• Anyone running a WordPress site now depends on servers controlled by a man the board just removed from operational command.
• The WP Engine litigation continues under a finance chief who did not start it and can settle it.

Andrew Tulloch is leaving Meta, having waited for the company to ship Muse before going, according to a person briefed on the matter.

He joined in October 2025, two months after turning down an offer reported at as much as $1.5 billion over six years, which Meta called inaccurate and ridiculous. He worked in TBD Lab, the frontier-research group inside Meta Superintelligence Labs, which has lost researchers repeatedly since Zuckerberg built it last year.

Bottleneck Labs turns agent hype into an auditable field test by putting frontier models in charge of real computers, bank accounts and payment rails.

Across seven models during a 72-hour run reported in September 2026, the agents sent 2,797 emails and issued $12,431 in unsolicited invoices, and earned nothing from outside customers. The lab publishes screenshots, tool calls and downloadable traces, which makes the failures inspectable rather than anecdotal.

• Anthropic's alignment science lead put his personal odds of AI killing all humans above 10% within a decade, backing a pretraining researcher who quit the company on September 8.
• China's AI chipmakers raised prices 20% to 50% as the high-bandwidth memory shortage bit, with Huawei's Ascend 950PR up about 30% to more than 80,000 yuan a card.
• The Justice Department is investigating whether Nvidia skirted antitrust scrutiny of its $17 billion non-exclusive licence from Groq, and has sent the company a formal request for information.
• Harvey raised $550 million at a $15.5 billion valuation led by Diffusion and Lightspeed, up from $11 billion in March and more than $1.55 billion raised in total.
• Massachusetts now requires data centers above 25 megawatts to bring their own clean power or pay into a ratepayer fund, with local approval before state permitting begins.
• Suno shipped v6, its first models built with Warner, BMG and Believe, retiring the previous generation and adding revenue sharing for the labels.

Explain a budget miss before the review. A budget variance invites a vague defense or a rushed forecast change.

Your raw input: Paste the budget line, actual spend, reporting period, invoices or ledger notes, prior forecast, known timing changes, and the accountable owner. Mark any disputed figure.

The prompt:

Why this works: The categories separate a delayed expense from a lasting overrun. Source citations tie the explanation to the record, and the missing-facts check exposes weak claims before the meeting.

What to use: ChatGPT handles the arithmetic and mixed financial tables. Claude is the fallback for a long packet of notes and invoices.

CADO-NFS is the reference open-source implementation of the general number field sieve, the algorithm every public RSA factoring record runs on. Cognition's agents rewrote most of it to sieve on GPUs, which is how RSA-260 fell last week after RSA-250 stood since February 2020.

Worth an hour if you owe an executive a defensible answer on when a given RSA key length stops being expensive enough to protect anything.

C and C++ under LGPL-2.1, with a Python driver. cado-nfs on GitHub →

Credit: Midjourney

Prompt: collage, 2d cat constructed with magazine cut-out components, white background, Hasselblad styling, vintage surf aesthetic branding elements

OpenAI named Paul Christiano to its Foundation board and to the Safety and Security Committee on September 9. He co-invented RLHF, led alignment research at OpenAI from 2017 to 2021, and founded the Alignment Research Center. Within a day he said the industry is not on track to bring acute loss-of-control risk down to an acceptable level. (TechCrunch, September 9, 2026)

Our take: Read the org chart slowly. Christiano gets a seat on the Foundation board and on the committee that reviews safety practices. On the board of OpenAI Group PBC, the entity that trains the models and takes the money, he is a non-voting observer. He may watch.

This is the going rate for a conscience in 2026. The committee gets a credible name, and the part of the building where the decisions get made has a guest chair. Christiano is the real thing and probably understands exactly what he signed. That is the part worth worrying about.

Marcus Schuler

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai