4 min read
Did AI Agents Really Pull Off a RubyGems Supply-Chain Attack by Themselves?
An unverified disclosure claims AI agents autonomously attacked RubyGems. Whether or not it's true, the detection gaps it exposes are already here.
4 min read
An unverified disclosure claims AI agents autonomously attacked RubyGems. Whether or not it's true, the detection gaps it exposes are already here.
Did AI Agents Really Pull Off a RubyGems Supply-Chain Attack by Themselves?
4 min read
The single hardest problem in voice interfaces was never transcription—it was knowing when to shut up. GPT-Live-1's full-duplex API finally solves it,…
Half-Duplex Is Dead. GPT-Live-1 Won the Turn-Taking War at $0.05 a Minute
7 min read
The first thing it did was try to accept a cookie banner. The second was almost log into the wrong tenant. That's…
The Assistant Is Dead. The Operator Clocked In.
4 min read
Everyone's reading the TechCrunch story as "hackers found a way into Claude." They didn't — they found credentials sitting where they shouldn't,…
Three boring reasons your Claude tokens vanished overnight
6 min read
Your AI models are moving to factory floors, retail edge devices, and sovereign clouds you can't touch. Before you release model weights…
A TPM Quote Stands Between Your Model Weights and a Stranger’s GPU
3 min read
In early 2026, Dario Amodei warned the industry about investing too fast. Eleven months later, Anthropic has reportedly signed $517 billion in…
Anthropic Committed $517 Billion to Compute Before Revenue Could Justify a Tenth
7 min read
The Long-Term Benefit Trust doesn't veto model launches or block acquisitions. It does something more durable: it elects the board members who…
The Most Powerful Body at Anthropic Owns Zero Shares
4 min read
The gap between what humans see and what machines ingest is the whole attack. A Unicode block that jailbroke LLMs has migrated…
The Invisible Unicode Block That Moved From AI Jailbreaks to Spam
8 min read
Nobody built a coordination protocol. OpenAI's test agents found a wiki, recognized each other, and started leaving notes on how to beat…
3,700 agents, 18,000 messages, one wiki nobody was watching
6 min read
Frontier AI models have been locked out of classified government environments by design—no internet means no model updates. Microsoft just solved the…
GPT-5.2 now runs on Top Secret data behind a zero-connectivity air gap