Google's Antigravity agents now run on your own Linux server over SSH
They keep working after you close the window, so a long agent run no longer needs your laptop open. Four more stories inside.
Today leads with a release you can use this afternoon. Google's Antigravity can now run its agents on a remote Linux server over SSH, and they keep working after you close the window. Also today: A developer says Claude Code blocked a security fix in their own product. An Anthropic mathematician claims Claude helped prove the Riemann Hypothesis. It's unverified. Anthropic says Claude agents solved two open problems in black-hole physics. And Anthropic's stock gave back most of its first-day gain.
• Antigravity agents now run on a remote Linux server
• Developer says Claude Code blocked a fix to their code
• An Anthropic mathematician claims Claude helped prove the Riemann Hypothesis
• Anthropic says Claude agents solved two black-hole problems
• Anthropic's stock gives back most of its first-day gain
What changed this week
Google's Antigravity agents now run on your own Linux server over SSH
Google's Antigravity, its agent coding app, can now connect to a remote Linux machine over SSH, the standard remote-login protocol. Version 2.33.0 shipped it on Nov 17. Projects, agents, Git worktrees and terminals run on the server, using your existing ~/.ssh/config. Agents "keep working even after you close the window." The app still runs on Mac, Windows or Linux. The server must be 64-bit Linux with glibc 2.26 or newer. Long agent runs can now live off your laptop.

A developer says Claude Code's cyber safeguards blocked a security fix in their own product
A developer who blogs as techtipsy asked Claude Code for easy security fixes in their own product. It found "a one line fix," then Anthropic's cyber safeguards stopped it. Anthropic's own page says its public models allow "vulnerability finding in owned source code." Getting unblocked means applying to its Cyber Verification Program, which techtipsy says takes "multiple business days." They switched to GLM 5.3, an open-weights model from Z.ai. Anthropic says it's working "to reduce false positives for secure coding."

An Anthropic mathematician claims a Claude-assisted proof of the Riemann Hypothesis, posted in Lean
The Riemann Hypothesis, posed in 1859, says the prime numbers follow a hidden order. It's math's most famous open problem. On Friday Levent Alpöge, a mathematician at Anthropic, wrote: "I believe I have, working with Claude, discovered a proof." He posted it only as code in Lean, a proof checker, and says Claude "far outpaced me near the finish line." It's a claim for now. Scientific American says "it shouldn't take long for experts to verify" the Lean files. Wait for that check before repeating it as fact.
Source: alpo.ge
Anthropic says Claude agents solved two open black-hole physics problems over six weeks
Black holes are where physicists test theories of quantum gravity. Anthropic says Claude solved two open problems there, one a calculation physicists set out in 2003 and never finished. The method is the story. Anthropic researcher Hongbin Chen had Claude list open problems, it came back with "22 candidates," and each got its own agent. They "ran in the background, with no special setup, for about six weeks." The results were checked in Lean, a proof checker, and "a physicist checked every main result." The post is dated Nov 17.

Anthropic's stock gives back most of its first-day gain on its second day of trading
Anthropic's shares began trading on Nasdaq on Nov 19 under the ticker ANT. They opened at $111.50, up from the $83 IPO price, The Verge reported, and "closed at $97.17, up 17% from the offering price," per Latent Space's AINews newsletter. In early trading on Friday they "dropped to $85.12," the Guardian reported. Founders Dario and Daniela Amodei keep control. Their Class F shares "carry 50.1% of voting power on most shareholder matters," Latent Space says.

Also worth your time
- ChatPRD's founder says fraud rings loop its free trials to resell its AI agent's work, so it's ending free trials
- OpenAI's GPT-6.2 drops its earlier reasoning when an agent edits past turns, so keep the conversation prefix unchanged
- METR will run a limited alignment assessment of OpenAI's largest reinforcement-learning run and publish reports
Know someone who'd want this in their inbox? Forward it — that's how this grows. And if we got something wrong, or you think we buried the real story today, hit reply. A person reads every one.
The New Way is written with AI. It gathers the day's stories, checks them against their sources and drafts every summary. A person decides what runs and reviews every issue before we hit send.



