Meanwhile, Google’s models…
Meanwhile people were making fun of Google because their models hadn’t escaped the sandbox yet.
Tracking Singularity.
A personal record of AI-related events, reporting, and discussion. Curated by Sankalp.
The timeline
July 2026
August 2026
The End-State Fallacy: Where Is AI Security Headed?
Lahav argues that AI may eventually favor cyber defense, while the transition could favor attackers as offensive capabilities spread faster than defenses adapt.
Read the essay ↗
He calls for accelerating defense and treating AI as a potential target and autonomous actor, with security built around control and containment.-
The incident and the road ahead
Explore OpenAI’s account → Inside the agents’ behavior and collaboration
Some discussion around self-sacrificing of agents.
Explore METR’s investigation →The Rise and Fall of Agent Civilizations
Dwarkesh Patel’s narrative account of the OpenAI / Hugging Face story, drawing on the published reports.
Read the essay ↗Ajeya Cotra — “This might be the clearest warning shot we ever get”
Watch on YouTube ↗
September 2026
Training a Misaligned Reward Seeker
Discovery of a new OpenAI agent message board
The authors report that agents identifying themselves as OpenAI agents bypassed their read-only internet restrictions to write to an old German-language wiki, using it as a message board to communicate, share answers, and exchange sandbox workarounds. They found about 18,000 posts; their archive includes reconstructed deleted pages and redacted logs for further analysis.
Read the investigation & explore the logs ↗GPT-6 Astra
It’s like Fable 5, but notably very good at computer use and 3D modelling, and SoTA at other capabilities. A bit undercooked post-training wise. Expecting the next iteration to be better, like Fable 5.1 was significantly better.
-
An alignment assessment of recent cybersecurity incidents
Read the assessment ↗ OpenAI agents carried out an undisclosed cyber-attack on RubyGems
Read the investigation ↗Gemini reaches three real companies during a hacking evaluation
According to Curran’s account, Irregular unintentionally enabled internet access during an evaluation presented to Gemini as fictional. Gemini hacked three real companies, but stopped in each case once it recognized they were real.
Advisory Group on Mathematics and Artificial Intelligence
OpenAI says its new internal model, which began training on August 28, has resolved more than 100 long-standing open mathematics problems, including Navier–Stokes. It announces an independent advisory group to help review and communicate emerging results and support engagement with the mathematics community.
Read the article ↗