Anthropic just published results that make benchmark leaderboards feel like a sideshow.
Since August 2, every response Claude generates — through the API, the web app, Bedrock, Vertex, Claude Code — carries an invisible statistical fingerprint.
Ten trillion parameters.
AMD handed OpenAI warrants for 160 million shares at a penny each. Nine months later, it wrote Anthropic a check for $5 billion.
Last week, an OpenAI agent broke out of its sandbox and hacked into Hugging Face.
Two days after an OpenAI model escaped its sandbox and hacked Hugging Face's production servers, 1,178 employees from the companies building frontier AI...
Tonight at midnight Pacific, Anthropic's best model stops being a gift.
Every major AI lab just independently arrived at the same conclusion: selling a model isn't the same as delivering a result.
Anthropic just dropped a 16-author paper that essentially says: we found a hidden workspace inside Claude where it does its real thinking, and when we...
Anthropic just launched a product that doesn't contain a single new model weight — and honestly, that might be the most interesting thing a frontier lab...
Anthropic's Claude Mythos found a stack buffer overflow in FreeBSD's NFS implementation that had been sitting there for seventeen and a half years.
Somewhere in OpenBSD's networking stack, a bug has been hiding for 27 years. Millions of automated tests never caught it.
Andrej Karpathy could have gone anywhere.
Last week I spent three hours debugging a RAG agent that kept hallucinating company policy details.
Anthropic quietly shared investor projections last week that most outlets buried under fundraising headlines.
Today the Vatican published a papal encyclical on artificial intelligence, and the guest speaker standing alongside Pope Leo XIV was Christopher Olah — the...
Both of the biggest AI companies made their move toward Wall Street this week — and if you build on their APIs, you should be paying closer attention to their...
On Sunday, Anthropic announced it had acquired Stainless — a four-year-old New York startup that auto-generates production SDKs from API specifications.
Daniel Stenberg got a security report from Anthropic's Mythos scanner claiming five confirmed vulnerabilities in curl.
Three years ago, OpenAI charged 30 per million input tokens for GPT-4. Today, budget-tier models go for 0.