Sponsored by
In Todayโs Issue:
๐ง Suspected hacks heat supermarket freezers across Europe
๐ณ DeepSeek publishes an agent built to leave you alone
๐ซ Anthropic confirms Claude's new rule against game clones
๐ Young mathematicians rethink their careers over AI
๐ Apple passes Nvidia as AI spending fears grow
โจ And more AI goodnessโฆ
โก The Signal
Someone is heating up Europe's supermarket freezers, and some of the doors were left unlocked.
Overnight, suspected attacks hit refrigeration systems in at least seven countries, three weeks after hundreds of French stores were hit. Danfoss, whose products were involved, says the Dutch systems affected were not on a secure network and still used a default password. No evidence ties any of this to AI, and the damage apparently needed none. That makes the next question uncomfortable: what happens to these machines when cheap models get good at this work, while the strongest ones are fenced off? Claude users are already reporting refusals on harmless reverse engineering, and SemiAnalysis's new GDPx benchmark says safety filters hold down top models' scores. Investors and young mathematicians are pricing in their own AI anxieties: Apple has passed Nvidia, and three in four surveyed early-career mathematicians see their academic odds falling.
All the best,
Kim Isenberg
(deepseek-ai on GitHub)
๐ณ DeepSeek's New Agent Is Built to Leave You Alone
DeepSeek has published an open-source prototype of a coding agent whose main design goal is to take up as little human attention as possible. Called dsh-superagent, it watches one GitHub repository, picks up work when an approved person asks, and answers in the same conversation, for example with a pull request. It confirms once, stays quiet while working, and writes again only when it has a result, needs a decision or cannot finish. DeepSeek labels it research code, not an official product, and it differs from the company's internal version. The repository passed 11,000 GitHub stars in about two days.
๐ tl;dr: DeepSeek released an open-source agent that takes coding requests in GitHub and only messages people when they need to act.
(anthropics/claude-code on GitHub)
๐ซ Anthropic Tells Claude to Stop Recreating Existing Games
Anthropic has confirmed that it told Claude Code to stop recreating specific existing games, and developers say the rule is catching legitimate work too. "We added two sentences to Claude Code's prompt on Opus 5.5 to steer it away from recreating someone else's specific game", Anthropic's Addy Osmani wrote on X, citing its ban on IP infringement. Original games, genre games and games from your own designs are meant to keep working. But a board-wargame developer says paid client projects stopped, and users also report refusals on routine reverse engineering, which that statement does not address.
๐ tl;dr: Claude Code is now told not to rebuild someone else's game, and some developers with legitimate game and security work say they are being refused too.
Survey authors Anna Brandenberger, Dora Woodruff and Hanna Mularczyk at MIT. (Jonathan Wiggs / The Boston Globe)
๐ Young Mathematicians Are Rethinking Their Careers Over AI
Three in four early-career mathematicians in a new survey say their odds of staying in academic math have dropped in recent months. Three MIT PhD students polled 268 graduate students and postdocs from almost 60 institutions; 75.5% reported the decline, and Business Insider counts 76% feeling less motivated. One wrote: "It is difficult to do research and go to seminars when it feels like none of the individual work will matter much in the future." The authors call it an informal temperature check, taken mostly before OpenAI's October 6 release of 700-plus AI-generated math results.
๐ tl;dr: Young mathematicians are telling senior academics that AI is already changing whether they stay in the field, and they want guidance.
I Almost Threw Out My Keurig. It Wasn't Broken.
Every cup from my Keurig started tasting weak, flat, stale โ so I unplugged it and stuck it in a closet. Then I learned the dirty secret of grocery store K-Cups: by the time a pod reaches your kitchen, it's usually spent months going from factory to warehouse to truck to shelf. The flavor oils are long gone. Some brands even over-roast to hide it.
The fix wasn't a new machine. It was fresh pods from Angelino's, a family-owned L.A. roastery that roasts, packs, and ships within 3โ5 days โ straight to your door, from 39ยข a cup. Over 1,000,000 happy customers. 50+ flavors, no subscription required, and 15% off your first order, applied automatically at checkout.
๐ฌ Watch This
Lex Fridman calls it "a kind of emergency episode". Three working mathematicians, MIT's Ankur Moitra, Steven Brunton of the University of Washington and Rutgers's Alex Kontorovich, go through what the recent AI claims on Navier-Stokes, the quasi-Riemann hypothesis and the unique games conjecture mean for their field. It picks up where today's survey stops, down to Moitra's advice for people starting out: "Pretend you already have tenure."
"You just start to realize that the way in which we built systems were kind of fundamentally insecure. We always knew it."
โ Greg Brockman, President and Co-Founder of OpenAI
Brockman argues that cyber capabilities available to everyone are rising, so defenders have limited time to use top models on old flaws. His company sells those models.
A new model name appeared on Anthropic's website, then vanished. The heading "Claude Claude Mythos 5.5" sat in the source code of its Mythos page in an October 31 archive capture, but not on October 23. Looks like they are getting ready to release their new Models.
Who Is Hacking Europe's Supermarket Freezers?
The Takeaway
๐ Suspected attacks disrupted fridges and freezers in at least seven European countries overnight, three weeks after hundreds of French supermarkets were hit the same way.
๐ Danfoss says its products were involved, and that the Dutch systems affected were not on a secure network and still used a default password.
๐ No group has claimed the attacks, and investigators have not said publicly how the attackers got in.
๐ Nothing links this to AI. The concern is what happens to machines this exposed as hacking skills get cheaper.
Someone has worked out how to ruin a supermarket's food from afar, and on Sunday night it apparently happened again. Fridges and freezers warmed up at Delhaize, Carrefour and Spar stores in Belgium and at SuperValu in Ireland, with similar reports from Sweden, Finland, Estonia and Luxembourg. Danfoss, the Danish maker of the cooling systems involved, counted at least 17 businesses hit in the Netherlands. One Swedish store said its freezers had been set to 50C. Three weeks ago, hundreds of French supermarkets lost stock the same way, and some had fires.
An employee cleans freezers that lost power at an ICA supermarket in Stockholm on November 2. (Fredrik Sandberg/TT News Agency/AFP/Getty)
The method looks almost insultingly simple. According to the CEO of Sabcobel, the Belgian company that operates the systems for Delhaize and other chains, attackers remotely switched units into defrost mode. Danfoss says the Dutch systems were not on a secure network and still used a default password, nearly three weeks after it told Dutch customers to disconnect them. Security researcher Kevin Beaumont called the French attack "a zero-skill thing to pull off". Fixing it is harder: Sabcobel's CEO says many units need physical changes, and "that requires investment".
Claroty researchers found a hidden login code built from the date in Danfoss's newer AK-SM 800A. Danfoss shipped a fix, and authorities have not said this route was used. (Claroty Team82)
So why does this belong in an AI newsletter? Nothing ties these attacks to AI. No group has claimed them, and investigators have not said publicly how the attackers got in. The reason is the size of the target: Shodan, a search engine for internet-connected devices, listed more than 4,500 Danfoss interfaces in October. CrowdStrike CEO George Kurtz, whose company sells the defense, argued on No Priors last month that AI gives a hacktivist the capabilities a nation-state had a few years ago. If a zero-skill method already empties freezers, that claim deserves a hard look from anyone running old machines on the open internet.
Why it matters: Old machines on the open internet have stopped being a quiet IT chore. Intermarchรฉ's owner expected French losses in the tens of millions of euros, and many of the fixes mean paying for hardware changes.
Sources:
๐ https://www.bloomberg.com/news/articles/2026-11-02/belgian-swedish-supermarkets-report-new-suspected-freezer-hack
๐ https://dutchnews.nl/2026/11/albert-heijn-sister-chain-hit-as-fridge-hacks-spread-in-europe
๐ https://www.cbsnews.com/news/europe-grocery-store-refrigerator-freezer-cyberattack-hack-food-wasted/
๐ https://www.theregister.com/a/5302784
๐ https://thehackernews.com/2026/10/attackers-heat-french-supermarket.html
The chart: SemiAnalysis's new GDPx benchmark gives AI agents 206 tasks in 11 fields it sees as the most valuable for frontier models, including chip design, quant trading, law and cybersecurity, then averages the scores. Claude Opus 5.5 leads at 12.0%, ahead of Gemini 4 Carbon at 10.9% and GPT-6 Astra at 9.6%. A task costs $51.41 on Opus and $16.29 on Carbon.
The lesson: Near the top, the price gap is now far bigger than the score gap: Carbon reaches about 90% of Opus's score at about a third of its cost per task, though it is not yet generally available. The limits in today's Claude story show up here too. SemiAnalysis says safety classifiers, the filters that block risky requests, hold down the scores of Opus and Astra, and it calls Meta's Muse Wonder Full the best available cyber model.
The caveat: A 12% top score is partly by design, since the tasks are meant to be hard, and SemiAnalysis picked the fields itself. In finance and law, it says, nearly every model scores 0.
๐ Apple Mostly Skipped the AI Spending Race. Now It's Worth More Than Nvidia.
โก Bottom line
Apple closed Friday at $5.03 trillion, passing Nvidia at $4.93 trillion after a rough month for AI-exposed stocks.
๐ก Why it matters
Investors are rewarding steady cash over AI ambition, as Alphabet and Amazon outspend their cash flow and borrow to build.
๐ What it means
The AI trade now faces a plain test: whether this spending earns enough, soon enough, to justify the debt.
Investors spent October asking who will actually get paid for the AI boom, and on Friday they gave an awkward answer: the giant that mostly stayed out of the spending race. Apple ended the day worth $5.03 trillion, ahead of Nvidia at $4.93 trillion, its first lead since a brief spell in late July. Neither stock had a good day. Apple fell 1.2% and Nvidia sank 4.2%.
The trigger was earnings. Alphabet, Meta, Amazon and Intel all disappointed, and the Nasdaq 100 lost 6.7% in October, its worst month since March 2025. CoreWeave and Nebius, which rent out AI computing, fell more than 30%. The worry is simple: AI chips and data centers cost so much that Alphabet and Amazon now have negative free cash flow, meaning investment outruns the cash their businesses bring in, and they are borrowing more just as interest rates rise.
Capital spending plans as of July 30, with Microsoft's as of April 29, before last week's updates. (Statista)
Apple is the contrast. It has kept capital spending tight, its shares are up 27% this year, more than double Nvidia's gain, and its free cash flow is projected to climb 38% to $142 billion for the fiscal year that ended in September.
Demand itself is holding up. Microsoft won applause last week for 48% sales growth at Azure, its cloud business. What has shifted is patience: investors want to see returns while the bills are still climbing. Apple's lead is only about $100 billion, and the last time it took the top spot, it held it only briefly.
Become an email marketing GURU.
Is your email strategy due for a tune-up? GURU Conference (powered by Constant Contact) is back Nov 12โ13, 100% free and virtual. You'll get email tactics you can steal the same day from top B2B and B2C marketers. Last year, 29,000+ marketers showed up. Save your free spot.
