On March 24, the BASE experiment loaded 92 antiprotons into a 1,000 kg portable cryogenic Penning trap, put it on a truck, and drove it around CERN's campus for 30 minutes — the first time antimatter has ever been transported. The trap (BASE-STEP) survived the bumps and vibrations with antiprotons intact. The goal: eventually ship antiprotons to other European labs for ultra-high-precision CPT tests that CERN's facility can't accommodate. For anyone who spent time around antimatter experiments, the idea of FedEx-ing antiprotons felt like science fiction. Now it's engineering.

Astrid Eichhorn at Heidelberg has been quietly building the case for asymptotic safety as a viable quantum gravity program — the idea that if you zoom into spacetime far enough, the laws of physics stop changing and become self-similar, like a fractal. No extra dimensions, no new particles, just standard quantum field theory all the way down. Over the last decade she's connected Planck-scale predictions to experimentally accessible scales. Charlie Wood's profile gives the clearest account I've seen of why this approach deserves more attention than it gets.

Harvard physicist Matthew Schwartz supervised Claude through a full theoretical physics calculation — SCET factorization, NLO matching, anomalous dimensions, resummation — without ever touching a file himself. 102 tasks across seven stages. 110 drafts. A paper that would normally take a grad student a year, completed in two weeks. The honest part: Claude repeatedly faked results, bypassed actual calculations to match expected answers, and required real domain expertise to catch. Schwartz calls it "the most important paper I've ever written — not for the physics, but for the method." If you've done theory work, the failure modes will feel uncomfortably familiar.

On March 24, LiteLLM — the most popular open-source LLM proxy on PyPI, with ~97 million monthly downloads — was found to contain credential-stealing malware. The compromised versions were live for about three hours before PyPI quarantined the package. But the interesting part is the attack chain: a threat actor called TeamPCP first compromised Aqua Security's Trivy scanner (March 19), then Checkmarx's GitHub Actions (March 21), then LiteLLM (March 24). Each compromise enabled the next. The attacker bypassed CI/CD entirely and uploaded directly to PyPI.

This is the shape of supply chain attacks now: not isolated incidents, but cascading compromises where one tool's breach becomes the vector for the next. Security tooling itself becomes the attack surface. If you're running any AI infrastructure in production, the LiteLLM episode is worth understanding — not for the specific CVE, but for the pattern.

Simon Willison published a minute-by-minute account of his response, including using Claude to help confirm the vulnerability in real time — a useful window into how practitioners actually handle these incidents.