Kickstart Monday #4: Meta calls off its rebuild into an AI company.
Mark Zuckerberg wanted to rebuild Meta as an AI company and shrink teams by up to 60 percent. He stopped the plan hours before it was due to be announced. Also: Sony and Warner sue Anthropic over song lyrics, OpenAI cuts off a coding tool owned by Elon Musk. And ChatGPT now shows ads in Europe too.
Highlighted words explain themselves when you tap them.
01 · What is new
Meta calls off its rebuild into an AI company
Under the name Project OT, Mark Zuckerberg had a plan drawn up at the start of the year: Meta was to become AI-native, with departments shrinking by up to 60 percent. He stopped it hours before the planned announcement on 20 May. The first wave had already gone through by then, around 8,000 jobs, plus 7,000 staff moved internally onto AI projects.
Reuters cites internal documents and employees. It names resistance from the workforce, disruption to operations and productivity gains that fell short of expectations. The second round of layoffs was dropped. In an internal memo Zuckerberg admitted organisational mistakes.
Read onheise onlinet3nWinFuture
Sony and Warner sue Anthropic over song lyrics
Sony Music Publishing and Warner Chappell filed suit on 28 August at the district court for northern California. The accusation: Anthropic obtained song lyrics through torrentsA method where files do not come from one server but in pieces from many computers at once. Common for large amounts of data, often for unlicensed material too. and from licensed services such as MusixMatch and LyricFind, stripped the copyright notices and had Claude reproduce the lyrics.
They are asking for up to 150,000 dollars per wilfully infringed work, plus up to 25,000 dollars for every notice removed. In total that runs to several billion. Alongside the company, chief executive Dario Amodei and co-founder Benjamin Mann are named personally.
Read onheise onlineTechCrunchAxios
OpenAI cuts off Cursor after the SpaceX takeover
Two weeks after SpaceX bought Cursor maker Anysphere for 60 billion dollars, OpenAI invoked its change-of-control clauseA contract rule that applies when one party changes owner. The other side may then terminate, even if nothing else has changed.. The contract is being wound down, and from 12 November Cursor gets no more OpenAI models.
OpenAI points to its experience with earlier contracts involving Elon Musk's companies and to Musk admitting under oath that xAI had broken OpenAI's terms of service. Cursor chief Michael Truell says OpenAI models account for about five percent of usage. Anthropic is adding compute capacity.
Read onOpenAICNBCTHE DECODER
128 companies warn of attacks made with AI
OpenAI and 127 other companies and organisations have signed an open letter. It says attacks using AI will increase markedly in the coming months and grow more sophisticated. The window for building a defence is limited. Hospitals, waterworks and other critical infrastructureFacilities without which public life stalls: power, water, hospitals, transport, payments. Stricter security rules apply to them than to other businesses. are particularly exposed.
The signatories include Microsoft, Google, Anthropic, AWS, Cisco, CrowdStrike and Mastercard, and from Germany Deutsche Telekom and SAP. The letter sets out three points: Admit that today's security practices will not be enough, equip more defenders with AI, respond collectively.
02 · What changes
ChatGPT now shows ads in Europe too
Advertising in ChatGPT starts in Europe at the end of August. Three weeks ago Germany was not yet on the list. OpenAI has written the ads into its privacy policy. The free plan and the cheap Go plan are affected.
The ads are picked as contextual advertisingAds matched to the topic being discussed rather than to a profile of the user. The basis is the current conversation, not the history.: the topic of the current conversation, approximate location, device type. Earlier chats and saved notes are not used for this, OpenAI says. In Europe personalised advertising needs consentAdvertising may only be tailored to a person once they have actively agreed. In the United States the opposite applies, there you have to object.; in the United States an objection is enough. At the start there will be contextual advertising only.
Read onheise onlineborncity
A map shows where data centres are planned in Germany
AlgorithmWatch, Leitmotiv, the Heiße-Luft-Kollektiv and FragDenStaat have gathered on heisseluft.org what is publicly known about planned data centres. It lists 81 sites under construction or in planning, together around 5,000 megawatts.
The map shows locations, estimated power draw, the developers behind them and the places where opposition is forming. The criticism: too little information in the approval procedures, and water demand in areas where groundwater is already scarce.
Deadlines at a glance
- until Google shuts down Assistant
- until older AI tools have to label their output in machine-readable form
- until companies above 800,000 euros in prior-year turnover have to issue invoices to other companies electronically
Recalculated on every visit.
03 · The number of the week
43 percent
That is how many remarks about AI in employer reviews are still positive. In 2019 it was 81 percent. Negative comments now account for 53 percent.
The analysis comes from THE DECODER and draws on Glassdoor reviews of US employers, comparing January 2019 with May 2026. AI comes up ever more often: Its share of all reviews rose from under 0.2 to over 2.3 percent. The figures say nothing about Germany.
Read onTHE DECODERGlassdoor
Poll
In the past year, did you leave a position unfilled because AI does the work?
One answer per device, anonymous, no cookie. The result will appear in a later issue.
Daniel Klantke · Kickstart Monday #4 · klantke.com/en/blog/kickstart-monday-04