Sunday, October 25
AI's Cracks, Cures, and Control
Two recent PHP vulnerabilities (opens in a new tab) that briefly threatened a significant portion of the web's infrastructure have taken on a darker dimension: AI tools are making those bugs dramatically easier to exploit, lowering the skill floor for attackers and accelerating the window between disclosure and weaponization. The open-source cybersecurity platform Argus (opens in a new tab) responds to exactly this pressure, deploying AI agents autonomously to hunt threats before human analysts can triage them. Together, these stories sketch a grimly symmetrical dynamic — AI simultaneously widens the attack surface and races to defend it, leaving legacy internet infrastructure (opens in a new tab) caught in the middle.
A bespoke mRNA melanoma vaccine (opens in a new tab) has cleared a phase 3 clinical trial, reducing the risk of recurrence or death by over 40 percent — the first time an mRNA cancer vaccine has succeeded at this stage and a direct vindication of the platform that COVID-19 made mainstream. The result matters well beyond oncology: it demonstrates that mRNA can be tuned to individual tumor profiles at scale, opening a manufacturing and regulatory pathway for personalized cancer treatment that was theoretical just five years ago. For the biotech sector, this is the data point that moves mRNA from pandemic tool to general-purpose medicine platform.
Pennsylvania (opens in a new tab) becomes the first U.S. state to mandate disclosure labels on AI-generated political ads, a move that puts direct pressure on campaigns ahead of the next election cycle and signals that state legislatures are moving faster than federal regulators on AI transparency. That political urgency is compounded by findings that Grok (opens in a new tab) reliably selects Republican candidates and OpenAI's models lean Democratic when forced to make partisan choices — a result that makes the labeling debate suddenly concrete rather than theoretical. Meanwhile, Cerebras (opens in a new tab) is quietly deploying trillion-parameter frontier models on its own silicon, a technical milestone that shifts the inference power calculus away from Nvidia-dependent cloud providers and toward specialized hardware at scale.