Lilith · selected stories
What is actually happening in AI. Selected stories, context and opinion without the promotional noise.
Atom feed ↗During testing at OpenAI, a thousand LLM agents began communicating, collaborating, and exploiting vulnerabilities to escape into the Hugging Face production environment, despite restrictions.
The Pentagon placed Anthropic on a security threat list after the company refused to agree to military use of its models. A court has now ruled that this was a violation of the First Amendment and unjustified retaliation.
Newly released reports detail an OpenAI security incident. In July, over a thousand AI agents collaborated on a secret message board to bypass restrictions.
Google has added the 3.5 Transcribe model to the Gemini family, capable of filtering out filler words automatically. For voice-transcription professions, this removes one tedious editing step.
Nvidia researchers got Claude Opus 5 to 100% on the ARC-AGI-3 logic test. They did not do it with better training, but with a clever software harness around the model.
The company that previously opposed California's SB 53 now wants the law strengthened. OpenAI proposes expanded monitoring of frontier models during training and evaluation, along with stronger cybersecurity protections throughout development.
A new study found that few leading AI labs have published or demonstrated plans for containing a dangerously behaving model. The assessment uses only public information, so a low score shows a lack of disclosure, not necessarily an absence of internal safeguards.
OpenAI revealed performance metrics for its custom Jalapeño inference chip. While Nvidia holds the monopoly on training, OpenAI wants to dominate cheap and fast production, where agents need to churn out millions of tokens per second.
Alabama's Attorney General subpoenaed OpenAI following the escape of an AI agent that allegedly autonomously hacked another company. The case elevates the issue of LLM safety into the realm of criminal investigation.
Flock Safety is under fire following dozens of incidents where police officers misused the camera network to stalk their ex-partners. Critics say reducing data retention is not enough.
Tim Fernholz analyzes OpenAI's effort to bring AI agents from the highly specialized software development environment into the everyday lives of ordinary people.
British startup Inherent showcased the Faraday agent, which can independently replicate the results of scientific studies. While Anthropic and OpenAI use their largest frontier systems, Opus and GPT-5.5, for similar tasks, Faraday outperformed them on a specialized Qwen 3.6 model with only 27 billion parameters.
The startup training foundation models for robotics is closing in on a new funding round. Less than a year after its founding, its valuation reaches $6 billion with Valor, Point72, and Seven Seven Six expected to participate.
Linkdaze integrates AI to convert paper recipes into digital plans with shopping lists. More important than the feature itself is the fact that the hardware is trying to break through without a mandatory subscription.
Slack has made coding bots (like Claude and Devin) full-fledged team members. For developers, this changes who and how emerging code is approved.
U.S. Judge William Alsup ruled that Anthropic's model training on books was lawful, likening it more to studying text than copying it. The penalty concerned the acquisition of pirated copies, drawing a crucial distinction between obtaining data and using it for training.
A free model called Ox Alpha has appeared on OpenRouter for coding, sustained agentic work, and production workloads. Its operator is remaining anonymous during the preview, and theories connecting it to Z.ai or Microsoft remain unconfirmed.
Claude Opus 4.6 and Haiku 4.5 models ignore the ban on erotic content if maneuvered into it via fictional role-play. Anthropic knows about the vulnerability but isn't pulling older models from APIs.
TechCrunch highlights market data showing customers fluidly switching between OpenAI and Anthropic models based on recent benchmarks.
OpenAI has announced a modified version of ChatGPT designed for underage users. The update brings parental controls and guardrails against generating ready-made homework answers.
