About
Why this website was created and how to use it
Tracking Singularity is a curated log of events in the AI world, covering capabilities, safety and alignment.
It was inspired by a run of events. After the OpenAI x Hugging Face hacking incident, reports started coming in almost every week of agents escaping their sandboxes, for one reason or another, and hacking websites to finish their tasks.
Around the same time, OpenAI and Anthropic started showing signs of recursive self-improvement (RSI) internally. They had managed to narrowly automate their research and training loop, which meant faster release cycles and an AI world moving faster than before. Fable 5.1, Astra and then Opus 5.5 were amazing releases because they no longer felt incremental. Labs started solving millennium problems like Navier–Stokes, and Anthropic found a new enzyme.
Then Jacob Coxon resigned from Anthropic. After three years of pretraining research at OpenAI and Anthropic, he wrote that neither company was taking enough care to make AI development safe, and that they were racing toward self-improving superintelligence.
The tweet made waves and set off a lot of discourse about “pacing the frontier”. Within days, Dario published We Must Pace the Frontier, and Elon, Sam and Dario all voiced agreement on pacing the frontier. When my father asked me about it, I walked him through everything in order, from the Hugging Face incident up to that tweet.
I often find myself telling people about these events, both the terminally online and the not so online. So I thought it would help to keep a log of them, along with my notes. And thus this website was born.
I’m also interested in these incidents because I want to work in alignment and AI safety in the near future, which is another reason I pay special attention to them.
How to use this site
Every entry on the timeline looks like this one.


- The site is built around articles and tweets. Each entry starts from a source, and the headline links to it.
- Reactions and my own thoughts live in Notes, linked under an entry.
- I often embed tweets on the timeline or in notes, so you can open them and read the replies and quote posts to dig further into the discussion.
- A chain connects related events across time, like the run of OpenAI agent incidents in the entry above. The chains come from the data itself, so if this site lasts, you may see the same data drawn in other, more artistic ways.
- To share an entry, click (or tap) its dot or date. That copies a link to the entry, and the link shows a preview card of it when you post it.
Support
If you’d like to support the upkeep of this site, I’m open to donations, GPU credits (like Modal) and grants (like Tinker). DM me on X or buy me a coffee.