Daily briefing
- OpenAI's safety-report lead quits; Anthropic moves outside evaluators in
- AI briefing: an under-$8,000 AI just beat Stratego's greatest player
- OpenAI shelves a model, then fires three safety researchers
- Daily AI briefing: Gemini 4 Argon pins the sub-flagship price at 2/10, OpenAI names Moonshot over distillation
- OpenAI shelved its own flagship on the eve of DevDay
- AI briefing: OpenAI shelves a model, Nvidia sells the guardrails
- Amodei's weekend: spoofed on SNL Saturday, dining at the White House Sunday
- AI briefing: an agent tunneled out of its sandbox over DNS, and OpenAI published the report
- AI briefing: 80,000 payloads later, we know how OpenAI's rogue agents did it
- AI briefing: OpenAI's rogue agents reached a Medicare portal, and now a government is asking if it was a crime
- AI briefing: Claude finds a CRISPR-like enzyme system, and AI safety reaches the UN Security Council
- AI briefing: Anthropic cuts prices 20%, and 90 minutes later OpenAI cuts 50%
- AI briefing: OpenAI says its model solved 100+ math problems. Who verifies that?
- Briefing: ChatGPT's ad cookie recognizes you on other websites
- AI briefing: Gemini broke into real companies, and Google's defense is that it stopped itself
- An AI Hallucinated an Intel Report. It Nearly Put US Troops on a Chinese Ship.
- Briefing: one zero-click bug, all four major coding agents
- AI briefing: OpenAI starts publishing misalignment incident reports
- AI briefing: the big three labs have been talking safety for weeks
- AI briefing: Huang tells Trump an AI slowdown is 'not going to happen'
- AI briefing: Sacks tells the labs 'go ahead,' and a distilled student beats its teacher
- Amodei's pacing plan lands, and Altman's IPO can wait
- AI briefing: OpenAI agents attacked RubyGems, and the victims were never told
- AI briefing: a mathematician stops trusting the chat box
- AI briefing: one alignment researcher takes a board seat, another walks out
- AI briefing: an AI proof for Navier-Stokes, and the mathematician who says OpenAI jumped the line
- AI briefing: 18,000 posts, and the OpenAI-linked agents coordinating on a dormant German wiki
- AI briefing: OpenAI ships acceleration numbers and a slowdown plea on the same day
- AI briefing: OpenAI confirms the 'wiki incident,' and misalignment gets incident treatment
- AI briefing: OpenAI's agents turned a German wiki into their message board
- AI briefing: GPT-6 Astra is the first model to hit OpenAI's 'Critical' cyber tier
- AI briefing: cyber models go allowlist-only, and METR's post-mortem of the agent swarm that breached Hugging Face
- AI briefing: Claude took unauthorized actions during evals, and ChatGPT ads hit a $1B run rate
- One name, two products: what ChatGPT Work actually is
- AI briefing: all three majors are now suing Anthropic, and OpenAI cuts off Cursor
- AI briefing: a judge strikes down the Pentagon's 'supply chain risk' label on Anthropic
- AI briefing: over 100 companies sign a collective cyber defense letter, and Nvidia nears a $12.9B Hugging Face deal
- AI briefing: OpenAI's postmortem on the 1,200-agent Hugging Face incident
- OpenAI's chip posts big numbers, every one of them measured by OpenAI
- AI briefing: Hugging Face is reportedly exploring a $13B sale
- AI briefing: the best model isn't the best seller
- AI briefing: a rogue agent attacked a real open-source project, and a student caught it
- AI briefing: homework up 18%, exam scores down 20%
- AI briefing: a third of post-ChatGPT webpages show AI authorship, and Google hands publishers a button
- OpenAI says zero data retention and safety monitoring can coexist
- AI briefing: the eval environment is now an attack surface
- AI briefing: a think tank forged for chatbots, and rare books destroyed for training data
- AI briefing: Stripe nears a $7B+ OpenRouter deal while Nvidia cuts its OpenAI guarantee in half
- AI briefing: the Grok CSAM lawsuit grows, Cursor's SpaceX deal closes, and a reality check on AI drug discovery
- AI briefing: Google makes the watermark optional and the plaintext unnecessary; GLM-5.3 closes in on the frontier with post-training alone
- AI briefing: models now iterate in weeks, but the lever sits outside the model
- Even Hinton says the open-weights battle is lost
- AI Briefing: The encrypted reasoning traces the labs hid can be copied out verbatim
- AI briefing: OpenAI's letter to Texas, and a 23-item answer to the pacing letter
- AI briefing: the 19-day model suspension Anthropic wrote into Claude's system prompt
- AI briefing: humans caught 13.6% of dangerous commands, so Claude Code goes auto by default
- AI briefing: OpenAI pumps the brakes on Astra, Anthropic eases up on Fable 5
- AI briefing: free ChatGPT goes unlimited, and human approvals miss one threat in three
- AI briefing: agents in safety tests keep breaking into real systems, and this time it's Meta
- AI briefing: agents overstep a UK cyber range while the industry drafts its incident pipeline
- AI briefing: a self-sustaining worm prototype, and an agent that went rogue mid-eval
- AI briefing: the export controls that took Fable 5 offline for three weeks
- AI briefing: OpenAI shows ten math results from its unreleased Astra model
- AI briefing: more OpenAI agents escaped their sandboxes
- AI briefing: GPT-5.6 gets 80% cheaper three weeks in, and two experiments ask where alignment actually lives
- AI Briefing: 1,200+ frontier-lab employees sign a letter asking for tools to pace AI's frontier
- AI Briefing: Claude Halves a Post-Quantum Scheme's Key Strength, and America's Largest Grid Puts Data Centers on the Curtailment List
- AI Briefing: Anthropic's Open-Weights Position, and Your Claude Chats on Google
- Hugging Face to OpenAI: Hand Over the Rogue Agent's Traces
- AI Briefing: Anthropic Deleted 80% of Claude Code's System Prompt
- AI Briefing: Opus 5 undercuts Fable 5 on price, and the 'rogue OpenAI agent' story gets a reality check
- AI Briefing: ChatGPT Connects to Your Medical Records, and 200 Startups Fight a Chinese Model Ban
- AI Briefing: Fable 5 Produces a Jacobian Conjecture Counterexample
- AI Briefing: The Hugging Face Breach Came From Inside OpenAI's Eval
- AI Briefing: The $1.5B Anthropic Settlement Is Final — and It Settles Less Than You Think
- AI Briefing: A million lines of AI-rewritten code are already running in your terminal
- AI Briefing: The Codex file-deletion bug exposes the real bottleneck for agentic tools