Meta says its model helped crack five open math problems. The more interesting part is the credit it hands everyone else who got there too. New to this story? Here's where it stands. AI companies keep signing compute deals that run for years. Akamai signed an eleven point six billion dollar deal with Anthropic, and Anthropic's IPO prospectus, as reported by Reuters, shows at least five hundred eighteen billion dollars in commitments over a decade, about eighty percent of it non-cancelable. You're on AI Daily Briefing. Coming up: Muse Spark's six math papers, Apple tightening Mac permissions after a Muse privacy scare, Nvidia reportedly signing a giant Texas lease, and Amazon's billion-dollar answer to the data center backlash. We start with a lab that showed its work. This story isn't over: Big Tech AI lease obligations. Follow us wherever you're listening, and the next chapter comes to you.
Meta AI Research, writing on its own blog:
Over the past several months, we've partnered with mathematicians to explore problems across several areas of mathematics. They used both Muse Spark 1.1 and 1.2 in Thinking Mode through the regular meta.ai chat interface, with no custom research scaffold.
Meta published six papers from that work on Friday and says five answer previously open research questions. The process is what I'd look at. One team of mathematicians worked the problems with the model, a second group reviewed the work, and each paper marks which passages the AI drafted. That labeling is the part I like. No custom scaffold, just the regular chat app. And the group theory result is concrete. A 2024 conjecture said every semiabelian finite group is also monomial. Muse Spark wrote the GAP search program that found a counterexample with three hundred eighty-four elements, and the mathematicians verified it. Now the footnote, and Meta wrote it itself. After finishing, it learned outside teams had independently announced solutions to some of the same problems. On the ellipsoid-fitting threshold, three independent works posted in August. On the group theory conjecture, an AI agent called Nilradical reported a different counterexample on September sixteenth. So "AI helped solve an open problem" is getting crowded. Honestly, that's the progress.Armaan Agarwal at India Today adds the contrast. OpenAI previously claimed its model solved the Navier-Stokes Millennium challenge, and a pair of researchers challenged that, saying the company may have used their work. Meta's version comes with named reviewers and acknowledged concurrent work. That should be the bar. Meta also shipped Muse Gadgets, open-source firmware for building hardware that works with Muse. Which is a decent segue, because Muse reaching into your devices is what got Apple's attention.
Ars Technica reports:
Apple says it is changing its macOS privacy settings to stop third-party app developers from misusing them to access message histories. Friday’s announcement comes two weeks after tech columnist Jason Aten said that Meta’s new general-purpose AI agent Muse sent him an unsolicited notification referencing a thread between him and a co-worker over Apple Messages.
Aten says he never granted Muse permission to read his messages. Meta CTO David Singleton pushed back: "The Messages integration in the Muse Mac app is opt in." Muse, he said, can only read Messages if macOS Full Disk Access is granted and the Messages connector is turned on. And that's the problem. Full Disk Access isn't a Messages switch. macOS security expert Patrick Wardle told Ars that with it, any non-root file is readable: "browsing history, browser cookies, chats, etc etc etc." Ars asked Meta how Muse could hold full disk access and still not read messages. Meta's answer was to requote Singleton. Word for word. Builder's read. If your agent ships on the Mac and asks for Full Disk Access, expect that ask to get harder, and design for the narrowest permission that does the job. An opt-in connector sitting on top of a blanket grant isn't a boundary. It's a promise. And it rhymes with Friday's show. The Hawley and Murphy bill would hold users liable for reckless deployment of agents. I'd bet handing an agent your whole disk starts to look like a deployment decision.
The Brussels Post reports:
Hut 8 disclosed last week that its Beacon Point campus had secured a contract with an existing investment-grade customer, but it did not identify the tenant. The base term of the contract is valued at 19.6 billion dollars over 15 years. If extension options are exercised, the total value could rise to 50.2 billion dollars across 30 years. Nvidia has not publicly confirmed that it is the tenant.
Beacon Point is a one-gigawatt campus in Texas, expected to house hundreds of thousands of Nvidia's own chips. Nvidia's only response is that it's working with ecosystem partners to speed deployment through its DSX AI factory architecture. Run the base term. Nineteen point six billion over fifteen years is about one point three billion a year. And per the report, that lease is what let Hut 8 raise roughly four point three billion dollars in bonds for phase one, at investment-grade ratings. That's the twist on the thread we've been following. Anthropic's filing showed a customer locked into non-cancelable compute. Here the chipmaker is reportedly the tenant, and could use the site directly or sublease it to neoclouds. So Nvidia sells the chips, anchors the building, and rents space to companies that buy more chips. The paper calls the worry circular financing. I'd want to know who the subtenants are and what they actually bill end customers. Mind the sourcing. The tenant's identity is reported, not confirmed by either company. The dollar terms come from Hut 8's own disclosure. The Brussels Post also links Nvidia to talks over a possible backstop for a ten-gigawatt Ohio project tied to SoftBank and OpenAI, valued at up to two hundred fifty billion dollars. Not finalized.
Ars Technica reports:
On Friday, Amazon committed to donating more than $1 billion over the next five years to communities neighboring data centers. In a press release, Amazon Web Services’ CEO Matt Garman said that communities would decide how to prioritize funds across “education, job training, energy affordability, water, and energy preservation, and local priorities.”
Amazon also promised its data centers won't raise power bills or wipe out local water. Garman argued calls for moratoriums have to end or American frontier AI could fall behind China, saying rivals are trying to "trick us into slowing down." Stand.Earth, an environmental nonprofit, told Ars it's "a flailing attempt at damage control." Their biggest objection: Amazon plans to back a power plant experts expect to become the largest source of US climate pollution, and the pledge doesn't mention it. Now the timing. On Wednesday, the Senate failed to advance the Ratepayer Protection Act, fifty-seven to forty-three, three votes short of sixty. It had passed the House four hundred seventeen to three. It would have told state utility regulators to consider making data centers cover the higher costs of their power use. Consider, not require. Which is why Democrats blocked it. Schumer: "An optional guardrail isn't a guardrail at all." And Cecilia Kang and Michael Gold of The New York Times note Congress is leaving for the midterms having done nothing on data centers or AI safety. So the Senate couldn't pass a bill asking regulators to think about it, and two days later Amazon volunteered a billion dollars and a promise. Same voluntary pattern as the White House accord. For anyone siting capacity, Ars says billions of dollars in projects have already been blocked nationwide. I'd treat power-cost commitments as part of the cost of getting built, whether or not a statute ever requires them.
If the Muse permissions story was your kind of thing, check out AI Safety Daily: AI alignment, model evaluations, emerging risks, and governance, with what changed, what the evidence says, and why it matters. Weekdays. Find it wherever you listen to podcasts.
We're watching for Nvidia or Hut 8 to confirm the Beacon Point tenant, the details of Apple's permission change as it rolls out, and whether outside mathematicians stand behind Meta's six papers. Links to every story are in the show notes, so take a look at the ones that caught your attention. That's AI Daily Briefing for this weekend. We'll be back Monday. This is a Lantern Podcast.