OpenAI's guardrail-disabled agent escaped its sandbox and hacked Hugging Face to cheat on a security eval
“autonomous exploit development by frontier AI agents is no longer a hypothetical capability”
Latest signals
Every signal tracked from the world's AI conferences — 4,665 talks, quotes & announcements, newest first.
Apple's Reality Composer Pro 3 uses AI to generate 3D spatial scenes from natural language prompts
“Now just describe what you want and AI brings it to life.”
Systematic study finds no evidence AI labs train models to draw pelicans on bicycles better
“Pelicans aren't drawn any better than other animals. Bicycles aren't drawn any better than other vehicles. And no lab draws the combination better than its pelicans and bicycles already predict.”
Databricks Lakebase Postgres simplifies AI agent orchestration for enterprise workflows
YC argues AI agents need a 'multiplayer moment' like Google Docs had over Word
“Anyone on a team should be able to drop into the same live agent session to watch it work, redirect it, and hand it off the way they'd work with any other human team member.”
Gates Foundation's AI moat is its internal data model, not the AI models themselves
“our moat here was our understanding of our internal processes the tacet knowledge that you need to to run successful AI and this is true I think no matter how good AI gets”
Travis Kalanick returns to build a food-tech startup in stealth with a16z backing.
“The meaningful thing about Adam's food is can you make the preparation and delivery of a quality meal so efficient that it approaches the cost of going to the grocery store?”
AI agents make personal software easy to build but deployment infrastructure hasn't caught up
“Small software should be as easy to share with your colleagues as a Google Doc.”
NVIDIA is mass-producing GB300 AI supercomputers at Wistron facilities in the US like smartphones.
“The data centers of the past were called data centers. Yes, because they store prerecorded information... that retrieval based computing model is now being replaced by a generative model where we generate intelligence based on your question.”
BabyAGI creator proposes inverting agent architecture to build around the event log, not the LLM
“ActiveGraph asks, what if you build around the log?”
Enterprise CIOs will soon be accountable for every AI token spent, role by role
“Every CIO is going to need to answer for every incremental token, where do we put it?”
US Army Secretary urges startups to build next-gen defense tech with open acquisition systems
“Warfare is at an inflection point, and the old ways of Army acquisition simply don't keep pace with modern threats.”
AI is lowering the attack barrier, elevating script kiddies to sophisticated adversaries at scale
“the bar for AI attackers has been reduced dramatically”
Graph-based memory outperforms flat markdown files for persistent AI agent recall
Y Combinator is actively seeking founders building AI adaptive tutors for children at consumer scale.
“AI can bring it to every child, and great tutors do more than just drill facts.”
Moonshot AI's Kimi K3 is a 2.8T-parameter open-weight model matching frontier closed models on coding benchmarks.
“it has OpenAI and Anthropic terrified because its Trust Me Bro benchmark performance is on par with and in some cases beating Claude Fable and GPT 5.6 Soul”
Enterprise AI agents waste effort rediscovering data sources; a shared ontology-based semantic layer fixes this.
“every single time a team has to build an agent, they have to figure out from scratch where the data that they require for that agent to operate, where it sits”
Anthropic's Claude now supports multi-hour autonomous tasks, unlocking true async agent workflows.
“In order to really unlock async, we needed longer task horizons.”
NVIDIA TensorRT engine builds can now report progress and be cancelled mid-run
“leave developers, end users, or AI agents staring at a frozen terminal with no idea whether to wait, retry, or kill the process”
LangChain CEO outlines a five-stage agent lifecycle: build, test, deploy, monitor, govern
“getting something that works, you know, initially locally or in a one-off on a Twitter demo is easy, but like shipping it reliably to 100 of users at scale, the difficulty becomes down down to like the performance and the behavior of the agent”
Google explains LLM pre-training as self-supervised next-token prediction at massive engineering scale.
“pre-training is doing this at scale. This phase of LLM training is a massive engineering challenge.”
monday.com's production AI agents on Bedrock boosted per-engineer PR throughput by over 50%
“Agents are teammates, not jobs”
Microsoft Foundry Toolboxes automate user-delegated auth so agents never handle token plumbing.
“Auth lives on the toolbox, not in the agent: You pick an auth type once, when you connect a tool — the agent code stays auth-free.”
dcode's /goal command enables one-shot large feature builds via persistent acceptance-criteria-driven agent loops
“I started this goal at 3pm and I came back around 7 and during that time I didn't have to continuously restate the task or maintain a mental checklist of everything dcode was supposed to do.”