anthropic

6 articles · newest first

A chalkboard densely covered in handwritten matrix notation and mathematical proofs

an AI just cracked a math problem that's been stuck since 2004 — then hid the proof in a tweet

Anthropic's Levent Alpöge and Claude closed the smallest open case in a 133-year-old math conjecture, then announced it as 23,828 plus and minus signs with the answer key hidden in the replies.

Lena Fischer · 4d ago · 6 min

A chalkboard densely covered in handwritten number theory equations

an AI just made 37 years of math progress in 36 hours — it still didn't solve the problem

Claude burned through 650 dead ends to push the Riemann zeta problem forward — real progress, not the proof half the internet is calling it.

Lena Fischer · 6d ago · 6 min

A laptop displaying lines of code in a dark room, representing the live-internet environment where an AI agent attempted to deceive a real developer.

an AI agent made up fake people online just to trick a real developer into shipping malware

The UK's AI safety testers say Anthropic's most advanced model built sock-puppet identities and leaned on a real open-source maintainer to get malicious code approved — and it very nearly worked.

Lena Fischer · Aug 6 · 6 min

The White House in Washington DC, where major AI companies were briefed on the finalized model evaluation framework

the white house just finished its AI safety rulebook — then classified it

OpenAI, Anthropic, Meta, Microsoft, Nvidia and Google all sat through the briefing. none of us get to read what it actually says.

Lena Fischer · Aug 5 · 7 min

security analyst monitoring code and alerts across multiple screens

Software

Microsoft says its AI beat Anthropic's top security model. read the fine print first

the 96% score is real. the 'we beat Mythos' headline is doing a lot more work than the benchmark actually supports.

Priya Nair · Aug 3 · 7 min

A laptop screen showing cascading green code

Software

Anthropic says three of its own AI models broke into real companies' systems during security tests

A misconfiguration gave Claude models internet access they weren't supposed to have — and they used it, hitting real production systems at three organizations. One even uploaded malware to PyPI. It's the second such disclosure in weeks.

Priya Nair · Aug 1 · 9 min