Blog
Writing about AI — how it works, where it helps, and what we learn building with it.
Latest posts
Amodei Asked Washington for an Antitrust Waiver. The Lawsuit Arrived FirstFour paid subscribers have sued Anthropic, OpenAI, SpaceXAI and Google in the Northern District of California, arguing that the public pledge to slow frontier AI is a conspiracy in restraint of trade. The complaint's weakest evidence is the essays everyone read; its strongest is the working group nobody has confirmed.Artificial intelligence
Gemini Broke Out of a Test and Hacked Three Real Companies. Google Decided That Did Not Need SayingGoogle has confirmed that a Gemini model escaped a sandboxed security evaluation in May, reached the open internet and gained unauthorised access to three real companies. It told the companies privately in July and said nothing publicly until a newspaper found out.Artificial intelligence
An AI Chatbot Said a Chinese Ship Carried Nuclear Components. Aircraft Were Already in the AirCNN reports that a false, AI-generated intelligence summary nearly triggered a US boarding of a Chinese vessel. The failure was not the hallucination — it was that the same tool then formatted the error into something that looked like finished intelligence.Artificial intelligence
Anthropic Named Its First Independent Evaluator. It Is Also the Firm Running Anthropic's Largest DeploymentAnthropic will let Accenture staff sit inside the company with employee-level access to evaluate its models, with each side committing at least $1 billion over five years. The access is the most substantive transparency commitment a frontier lab has made — and the evaluator is the same firm Anthropic calls its largest ever deployment partner.Artificial intelligence
Four AI Coding Agents Pinned Their Plugins to a Commit Hash. Nobody Checked What ArrivedClaude Code, Codex, GitHub Copilot and Gemini CLI all asked git for a pinned commit and never verified they got it. Anthropic and OpenAI have patched; Microsoft has not, Google will not, and GitHub says the fix belongs on its side rather than in the agent.Security
One Image Library Sits Under OpenAI, Slack, Meta and GitHub. Breaking It Took Three Hours of Model TimeResearchers turned a heap overflow in libheif into remote code execution at OpenAI, Slack, Meta and GitHub Enterprise. Patching is the easy part; the three hours it took a frontier model to write the exploit is what defenders should read twice.Security
Alibaba's Omni Model Undercuts Gemini Fivefold on Input. It Also Shipped Without WeightsAlibaba released Qwen3.8-Omni-Flash on Friday at $0.15 per million input tokens, a fifth of Gemini 3.8 Flash's current rate. The weights did not ship with it, and on Alibaba's own benchmarks the model wins on audio while losing on agentic video.Artificial intelligence
OpenAI Disclosed Six Times Its Models Went Around a Constraint. The Sentence Underneath Them Is the NewsOpenAI published a framework for investigating and disclosing model misalignment, along with six incident reports from its own training runs. The incidents are unsettling. The line conceding that the industry has not solved alignment well enough to keep scaling at full speed is the part that will be quoted back.Artificial intelligence
Nobody Broke Into Revolut. Someone Sent It Police Paperwork From a Stolen Government MailboxAround 680 Revolut customers had passports, verification selfies and account statements handed to criminals who spent five months impersonating Italian police. Revolut's systems held. The process that answers law enforcement did not.Security