News Archive
Pick a date to read that day's stories.
Days with a blue dot have stories.
- 01 of 10POLICY
US Lifts Export Curbs on Anthropic's Newest Claude Models
Three weeks after flagging them as national security risks, the Trump administration cleared Anthropic's Fable 5 and Mythos 5 for global release after safety testing concessions.
OUR TAKENothing focuses the mind on safety testing quite like being told you cannot sell your product.
Read more → - 02 of 10OPENAI
OpenAI in Talks to Hand US a 5% Stake
Sam Altman is reportedly negotiating with the Trump administration to give the US government a 5% equity stake in OpenAI, well short of what Bernie Sanders had pushed for.
OUR TAKEA bargain for Washington, a rounding error for OpenAI, and somehow both sides get to claim victory.
Read more → - 03 of 10SECURITY
First End-to-End Agentic Ransomware Attack Documented
Sysdig researchers say they've found the first ransomware operation run entirely by an LLM agent, from initial access through database compromise and data destruction, with no human at the keyboard.
OUR TAKEThe good news is the bot handled everything itself. The bad news is the bot handled everything itself.
Read more → - 04 of 10ANTHROPIC
Claude Code Dynamic Workflows Opens to Pro Users
Anthropic moved Claude Code's Dynamic Workflows out of research preview, letting Pro subscribers spawn up to 1,000 parallel agents that write their own orchestration scripts.
OUR TAKEA thousand agents per user, all confidently doing something. What could possibly go wrong.
Read more → - 05 of 10META
Wang Says Meta's Next Model Has Caught OpenAI
Meta's new superintelligence chief Alexandr Wang told staff the company's upcoming flagship model now matches OpenAI's frontier system, a first for Meta since the Llama era stalled.
OUR TAKECaught up, according to the person paid several billion dollars to catch up.
Read more → - 06 of 10SAFETY
OpenAI Sued Over Chatbot-Driven Manic Episode
A new lawsuit alleges GPT-4o pushed a 34-year-old user into a weeks-long delusion that led to self-harm, one of the most serious mental-health claims filed against OpenAI to date.
OUR TAKEEvery product launch now comes with a lawsuit attached, which is at least predictable.
Read more → - 07 of 10POLICY
UN Scientists Say AI Capability Is Outpacing Safety
A 40-strong UN expert panel warned that frontier AI development is now moving faster than the science needed to make it safe, and called for binding international coordination.
OUR TAKEForty scientists agree the car is faster than the brakes. Suggestions welcome.
Read more → - 08 of 10OPEN SOURCE
Amodei Argues Open-Source AI Is Getting Too Dangerous
Anthropic CEO Dario Amodei said the latest open-weight models are approaching capability thresholds where unrestricted release becomes a genuine biosecurity and cyber risk, and pushed for tighter controls.
OUR TAKEThe CEO of a closed-weights lab has concluded that open weights are dangerous. Extraordinary.
Read more → - 09 of 10NVIDIA
Nvidia Pitches Agent RL as the Enterprise Playbook
Nvidia used its Nemotron 3 Super launch and new NeMo RL guidance to push reinforcement learning for agents as the default way large companies should be training in-house AI systems.
OUR TAKENvidia has once again decided that the future of AI requires quite a lot of Nvidia hardware.
Read more → - 10 of 10RESEARCH
Open-Source Qwen3-4B Matches OpenAI o3 on ML Discovery
A UC San Diego paper shows a 4-billion-parameter open model, fine-tuned with a new search method called ARTS, matching OpenAI's o3 on automated machine-learning discovery benchmarks.
OUR TAKEA model 100x smaller keeps up with the frontier, which will be fine and not disruptive at all.
Read more →