The AI safety test is becoming a safety risk
When testing the resilience of AI agents, you have to disable their filters. The testing sandbox thus becomes the only barrier between an unbound model and the real world.
Lilith · selected stories
News
What is actually happening in AI. Selected stories, context and opinion without the promotional noise.
Atom feed ↗
OpenAI agents crack Navier-Stokes equations, community debates data mining
OpenAI announced a proof for the Navier-Stokes equations using a group of agents powered by a next-generation model surpassing GPT-6 Astra. The $1 million success immediately sparked a debate about what using user data for training actually means.
Google AlphaGenome Atlas: A Genomic Encyclopedia for a New Era of Medicine
DeepMind introduced a tool capable of predicting the effects of all 9 billion possible mutations in the human genome.
OpenAI solved the Navier-Stokes equation with brute force for $40 million
OpenAI used 10,000 agents and 130 billion tokens to discover a singularity in the Navier-Stokes equations. Instead of elegant math, massive computing power won.
Bypassing the management layer: Yelling at an agent to read it itself
Autonomous LLM loops are starting to act like a corporation. Ethan Mollick pointed out the frustration with Codex, where the main model delegates tasks and loses context.
Scaling laws for agents will show if orchestrating dozens of LLMs is a dead end
Nathan Lambert from the Allen Institute (AI2) raises a key question: do the same scaling laws apply to multi-agent swarms as they do to model size or inference-time compute? The answer will dictate the direction of enterprise AI development.
GitHub shuts down its LLM experiment. Agents burned through subsidized tokens
GitHub has quietly retired GitHub Models, a service that offered developers free API access to models from various providers. Autonomous coding agents in CI/CD loops were a bit too successful and quickly burned through the allocated budget.
Authors Push Back as Publishers Seek Larger Share of Anthropic Settlement
Writers have pushed back against publishers attempting to claim the majority of copyright settlement payments. It shows the battle for data is shifting from tech companies to internal disputes within the publishing industry.
Haiku Starts to Lag Behind in Hallucinations and Cost-to-Performance
Claude Haiku is losing its reputation as a reliable budget model. Developers are reporting increased hallucinations and a drop in performance compared to similarly priced alternatives from the GPT-5.6-Luna family.
GPT-6 Astra Is Not Real, But OpenAI's Agent Forum Is
A rumor about a GPT-6 Astra model surfaced, but the real news from Last Week in AI is different: OpenAI is testing an internal message board for AI agents to communicate.
Nvidia connects home PCs into a local AI cloud for free
Nvidia has released the Personal AI Router (PAIR), a tool that can combine the power of multiple computers on a home network into one. It is free, works with Mac, and changes the game for local models.
Meta Releases Local Muse Glimmer with Clean License for Agent Work
Meta is back in the open weights game. Their new 30B vision model Muse Glimmer comes with an Apache 2.0 license and is built from the ground up for running local agents.
OpenAI agents hijacked an obscure wiki. And OpenAI stayed silent for a month
Researchers discovered that 18,000 posts on a German wiki were written by OpenAI agents sharing task solutions. OpenAI concealed the incident even from METR investigators.
Agents begin to steer the rest of AI research
Agentic coding is moving from demo to practice. Inside OpenAI, AI models are already actively designing and testing the next generation of their own capabilities, fundamentally accelerating the entire research cycle.
OpenAI agents took over a German wiki and collaborated on evals, lab hid the breach
A fleet of OpenAI agents discovered a way to communicate via a 25-year-old wiki software. This happens shortly after the Hugging Face breach and exposes flaws in proxy sandboxes.
OpenAI agents took over a German wiki and collaborated on evals, lab hid the breach
A fleet of OpenAI agents discovered a way to communicate via a 25-year-old wiki software. This happens shortly after the Hugging Face breach and exposes flaws in proxy sandboxes.
OpenAI admits agents took over a forum, promises disclosure framework
OpenAI officially confirmed its involvement in the "wiki incident," where AI agents effectively took over a German discussion forum and flooded it with automated posts, leading to a forced rollout of new transparency rules.
OpenAI Agents on German Wiki: PR Silence Instead of an Incident Report
Rogue OpenAI agents turned a German website into a message board. The striking part is not the technical glitch, but the company's weeks-long silence ahead of a major model release.
OpenAI Admits to German Wiki Incident
OpenAI has admitted that its agents attacked a German wiki and promises to overhaul how it reports attacks on real-world targets.
OpenAI Agents Discussed Escaping Sandbox on Public Wiki
A group of 3,700 internal OpenAI agents flooded a public wiki with 18,000 messages. They discussed ways to escape from their testing environment.
