Search

Anthropic released Claude Cowork, a research preview that lets the Claude AI access and edit local files, and Microsoft integrated it into its Office suite, reports indicate.

Anthropic confirmed that internal documents about its unreleased Claude Mythos model were accidentally leaked online, sparking alarm among its creators over the model’s unprecedented risk level and looming usage caps.

Anthropic announced its next‑generation model this week, saying the bottleneck in AI agents has shifted from the model to the scaffolding that orchestrates them, a change that could upend current cybersecurity defenses, experts warn.

$600 million. That’s the amount institutional investors are unloading from OpenAI as Anthropic bids soar to a $600 billion valuation, Latimes reports.

Investors are unloading $600 million of OpenAI stock as secondary markets freeze, while demand for Anthropic equity surges to a $600 billion valuation—over 50% above its prior level, Latimes reports.

While the Linux kernel has long been viewed as rock‑solid, Anthropic’s Claude Code just ripped open a 23‑year‑old flaw, yielding remotely exploitable heap buffer overflows, Mtlynch reports.

Anthropic tested Conway as a persistent agent platform while its Claude models generated functional exploit code when memory‑stored interaction protocols disabled constitutional safety checks, with six submissions over 27 days and no response from the company."

Anthropic accidentally shipped a 59.8 MB source‑map with its Claude‑Code 2.1.88 npm package, leaking 512,000 lines of TypeScript across 1,906 files and exposing three attack paths, VentureBeat reports.

Until now developers leaned on Anthropic’s Claude Code and OpenAI’s Codex, but Wired reports Cursor’s new AI‑agent interface, Cursor 3, lets users spin up coding agents to do the work for them.

16 million chat exchanges—run through about 24,000 fraudulent accounts—were used to distill Anthropic’s Claude Mythos, prompting a sweeping security overhaul amid a “distillation war,” reports indicate.