Anthropic Pledges 5% of Its Shares to the Public as Microsoft Quits a Grid-Starved Essex Data Center and MiniMax Ships M3.1 Flash
The weekend’s biggest AI stories were about money, power, and people rather than new frontier models. Anthropic committed 5% of its shares to a new Public Benefit Fund, days after publishing data showing its share of junior technical hires has fallen sharply and its staff expect AI to do most of their jobs by 2030. Microsoft walked away from Nscale’s flagship Essex data center after Nscale was told the grid cannot power it until the 2030s. MiniMax shipped the low-cost M3.1 Flash, Disney agreed to hire most of bankrupt Character.AI’s staff, and Binance launched a trading agent that cannot move money without its user.
Anthropic Commits 5% of Its Shares to a New Public Benefit Fund
Anthropic said on Sunday, October 18, that it will put a slice of the company itself toward the public good. In Anthropic’s announcement of the Anthropic Public Benefit Fund, the company says that, subject to stockholder approval, 5% of its shares will be committed to the new fund immediately, and the board may add another 2% for every 50% increase in the company’s value above a set baseline. The fund is a wholly owned LLC rather than a 501(c)(3), which Anthropic says gives it room to move between forms of giving, from large donations to traditional nonprofits to direct distributions to specific populations. Labor market disruption is the example it leads with: Anthropic has already committed $200 million to studying and addressing AI’s effect on workers and says “bolder approaches may be needed.”
"None of the Anthropic equity held by the LLC will be used for lobbying or political giving, in the US or abroad." — Anthropic
Counting the co-founders’ founding pledge to donate 80% of their equity and an employee equity donation match that about half of staff use, Anthropic says roughly 20% of its shares will now be dedicated to the public good. Anthropic’s official post announcing the fund on X went up at 11:07 AM PT and repeats those terms. CNBC’s report on the Anthropic Public Benefit Fund notes that the valuation baseline has not been disclosed, that Anthropic told investors its third-quarter revenue stands at $16.8 billion, and that its initial public offering is reportedly slated for November. IPO adviser Lise Buyer of Class V Group told CNBC the timing is both necessary before an IPO and “beneficial from an image perspective.” TechCrunch’s coverage of the Anthropic fund points out the open question: who decides which causes get funded. Anthropic says it will publish the structure, governance, and valuation milestones in the coming months, along with a public benefit report.
An escalating equity pledge is a credible signal only once the governance is public. Until Anthropic names who controls the fund, how the baseline is set, and how payouts are decided, treat this as a promise, not a program. For workforce planners, the useful detail is the explicit focus on labor disruption, which could fund retraining.
Anthropic’s Own Data Shows Fewer Junior Hires and Staff Who Expect AI to Do Most of Their Work
The fund lands days after Anthropic published two studies on AI and work, both dated October 16. Anthropic’s report on how employment is changing at Anthropic says the share of junior-level employees in its technical departments slipped from 21% to 14% since early 2025, while the share of junior technical hires fell from 27% to 9%. Technical departments shrank from 61% to 56% of the workforce even as technical hiring grew in absolute terms. By August, agents that employees ran interactively logged six times as many hours as their human handlers, and code commits per engineer were running at five times the rate of six months earlier. The company says it has set up an early warning system to track jobs at risk of disruption and has prioritized internal redeployments.
"This acceleration makes human attention the bottleneck for agent output." — Anthropic
TechCrunch’s report on Anthropic’s less junior, more meeting-heavy workforce adds that calendar data show meeting time per employee up 20% from a year ago, and up 16% for technical contributors. The companion study, Anthropic’s AI Expectations Gap survey of US adults and Anthropic staff, compares 10,980 US adults polled by Morning Consult in August with 1,334 Anthropic staff. On average, staff expect AI to be able to do 75% of their current work by 2030, against 28% for US workers, and 91% of staff expect AI to make a Nobel-level discovery by 2030, against 39% of adults. Run through Anthropic’s economic model, the median staff member’s answers imply 17.1% GDP growth in 2030, versus 5.4% for the median adult.
Jack Clark, Anthropic’s co-founder and head of public benefit, took that view to the IMF and World Bank annual meetings in Bangkok on Saturday. Reuters’ report on Jack Clark’s IMF remarks, via The Manila Times, says one of Anthropic’s internal models suggests that under certain conditions US output could rise to about $44 trillion by 2030 from just over $30 trillion, which Clark conceded “sounds very outlandish, I know.” He said AI is costing jobs in call centers and customer service, and that the window for safeguards is now “a small number of years.”
"We should expect that in the coming years, the whole pace of progress of not just AI but science as a whole will speed up, and this will feel confusing and disorienting, and will bring with it great wealth and potentially great peril" — Jack Clark, Anthropic, via Reuters
Anthropic is the most AI-saturated employer on record, so its junior hiring data is an early signal, not a forecast for everyone. Still, leaders should ask now where entry-level talent will learn judgment if agents absorb the routine work, and should measure reviewer attention as a real capacity limit when they scale agent output.
Microsoft Walks Away From Nscale’s Grid-Starved Essex Data Center
Britain’s flagship AI data center has lost its anchor tenant. BBC News’ report on Microsoft pulling out of Nscale’s Essex AI data centre says Microsoft, which had said it would fund a supercomputer at Nscale’s planned £2 billion site in Loughton, has ended its involvement after Nscale was told the site’s power connection, due at the end of this year, will not come until the 2030s. The Telegraph first reported the withdrawal on Sunday. AFP’s report on Microsoft confirming its exit from the Nscale deal says Microsoft confirmed on Monday that “this decision relates to a single site” and that it remains “committed to our partnership with Nscale.”
"We remain committed to both the Loughton project and the UK and we are exploring alternative power sources for the site, in addition to other expansion sites in the UK" — Nscale, via BBC News
The Register’s analysis of Microsoft abandoning the UK AI supercomputer recalls that the September 2025 plan was to put 23,040 Nvidia GB300 GPUs into a 50MW facility, expandable to 90MW, with first chips in early 2027. UK Power Networks told The Register that many data centers can connect to its distribution network but are waiting on national transmission works it does not own. The Register adds that demand for new grid connections surged from 41GW to 125GW in under a year, with data centers accounting for at least 80GW, prompting Ofgem to propose charging developers for connection requests. The Guardian’s report on Microsoft dropping the Essex datacentre says Microsoft was the only publicly announced customer for the site and remains on board for an Nscale project in Norway. The timing is awkward: SiliconANGLE’s report on the risk to Nscale’s IPO plans notes that Nscale has applied to list on the New York Stock Exchange under the ticker NSCL, and that Microsoft and Anthropic account for about 85% of its contract value, according to its IPO filing.
Power, not chips, is now the binding limit on AI capacity, and neoclouds carry that risk with very few customers to spread it across. Any enterprise counting on dedicated regional GPU capacity should ask providers for a confirmed grid connection date, the backup generation plan, and what happens to its contract if a site slips by years.
MiniMax Launches M3.1 Flash and Reveals It Was the Anonymous “Space Bunny”
Chinese lab MiniMax launched M3.1 Flash on Monday, October 19, and confirmed a leaderboard mystery. MiniMax’s launch post for M3.1 Flash says Space Bunny, which processed more than 200 trillion tokens while operating anonymously, was an early version of the model. MiniMax claims nearly three times faster online inference and 70% lower usage costs on agentic tasks than its predecessor, with flat pricing of $0.15 per million input tokens, $0.60 per million output tokens, and $0.015 per million cached input tokens. Its self-reported scores include 73.7% on DeepSWE v1.1, 61.7 on OSWorld V2, 29.6 on Agents’ Last Exam, 76.2 on BankerToolBench, and 82% on CyberGym, and it says output speeds can top 200 tokens per second.
"M3.1 Flash is now available through the MiniMax API, M Plan, and MiniMax Code. Model weights will be released later." — MiniMax
MiniMax’s official M3.1 Flash launch post on X, published at 4:56 AM PT, gives a date for those weights: October 23. It also puts the cost at one-third of M3’s, and adds a 91% score on the BabyVision visual reasoning benchmark. MiniMax says a Pro model that pushes further on capability will arrive later.
At fifteen cents per million input tokens with near-frontier agent scores, M3.1 Flash will pressure every mid-tier API price, but all of those benchmarks are MiniMax’s own. Teams should run it on their own agent tasks, compare cost per finished job, and wait for the October 23 weights before planning any self-hosted deployment.
Disney Agrees to Hire Most of Bankrupt Character.AI’s Staff
One of the best-known consumer chatbot startups is ending up as a talent deal. TechCrunch’s report on Disney hiring the majority of Character.AI’s workforce says a legal filing, first spotted by Bloomberg, shows Disney agreeing to make employment offers to at least 60 Character.AI employees, the majority of the workforce, at an estimated $60 million in annual compensation, and that the “vast majority” of those offered jobs accepted. Character.AI filed for bankruptcy earlier in the week. Disney had already agreed to buy some of its assets for $20 million in a stalking horse bid, so a rival could still bid more. Character.AI’s former CEO, Karandeep Anand, recently became Disney’s first CTO. The filing says that despite a “large and engaged user base,” the company was hobbled by high costs and “potential litigation-related liabilities and regulatory engagement.” TechCrunch notes that Character.AI was sued by the mother of a 14-year-old boy who said her son became obsessed with one of its chatbots before his suicide, and that it faces government investigations over child safety.
Standalone companion apps carry high running costs and growing child-safety liability, and in this case the team proved easier to sell than the product. Brands planning character or companion experiences should budget for age assurance, safety review, and legal exposure from day one, since the filing names exactly those burdens.
Binance Launches an AI Trading Agent That Cannot Move Funds on Its Own
Agents are reaching money, with guardrails attached. Binance’s press release launching Binance AI Pro, issued Monday, describes a finance-focused agent that turns plain-language trading ideas into automated strategies shown as an editable visual workflow. Every strategy can be paper traded with virtual funds first. Going live needs a separate review-and-approve step and runs in an isolated AI Strategy Account that only the user can fund by manual transfer, and that cannot withdraw to external addresses. The agent cannot move funds by itself, and strategies can be paused or revoked at any time. Binance says the design follows the “bounded autonomy” principles in its own policy paper, under which agents analyze and recommend while consequential actions require explicit human approval. The free Standard Plan includes 250,000 monthly credits; the 19.99 USDC-a-month Premium Plan raises that to 10 million and adds frontier models where available. Backtesting is not in the first release.
"Users describe the idea, review the workflow, test it with paper trading, and approve deployment themselves." — Jeff Li, Binance
Binance’s design is a usable template for any agent that touches money: a sandboxed sub-account, user-only funding, no external withdrawals, and human approval before going live. Builders should copy those controls, while users should remember that paper trading is not backtesting and that volatile markets can still lose real money fast.
The AI boom is now testing its own foundations. Anthropic is pledging equity to cushion the labor disruption its own data shows arriving first among junior engineers, while its staff expect far faster change than the public. Microsoft’s retreat from Essex shows that grid power, not demand, sets the pace of new capacity. Cheap models like M3.1 Flash keep pushing prices down, a once-popular chatbot startup is being broken up for talent, and agents that touch money are shipping with human approval built in. Plan for faster automation, slower infrastructure, and tighter controls.
Need help navigating AI for your business?
Our team turns these developments into actionable strategy.
Contact SEN-X →