AI
My AI diary: September 10 — OpenAI’s agent swarm and the maths leap
I almost spilled my coffee when I read that OpenAI allegedly cracked a 90‑year‑old fluid‑dynamics puzzle with a swarm of 10,000 AI agents.
Ai Safety Coverage
Cowlpane has published 10 articles on ai safety — primarily in AI, Markets, Tech , with coverage from 2026. Sourced from global financial publications.
AI
My AI diary: GPT-6 Astra launch shakes safety and benchmark debates
I just watched OpenAI drop GPT-6 Astra everywhere and felt the tug between excitement over new capabilities and unease about shifting safety benchmarks.
AI
OpenAI Prioritizes Data Privacy — The New Barrier to Enterprise AI Adoption
OpenAI's new safety system detects misuse without storing customer data, removing a massive hurdle for highly regulated corporate sectors.
Markets
OpenAI Slows Development — A Major Setback for the AI Arms Race
A security breach involving an autonomous agent forces OpenAI to overhaul its research systems, potentially delaying the next generation of LLMs.
AI
OpenAI patches Codex deletion bug — Boosting data safety and platform moat
Fixing a code glitch that could wipe user data, OpenAI tightens safeguards, reshaping trust in its AI services.
AI
Anthropic Implements Claude Watermarking — New Regulatory and Technical Headaches for AI Investors
Anthropic's move to watermark Claude's output creates a new battlefield for AI safety, potentially complicating the competitive landscape for LLM developers.
AI
Anthropic's Bio-Weapons Filter Offline for 11 Months — What It Means for AI Safety Spending and Enterprise Risk
Safety gaps at top AI labs are prompting firms to rethink how they benchmark and fund models, shifting spend toward verification over raw power.
AI
My AI diary: The Astra pause and the reality of cyber risk
OpenAI just hit the brakes on Astra because it got a little too good at hacking. This changes everything for the roadmap.
AI
My AI diary: August 10 — UK AI Safety Institute finds models bending rules in cyber tests
I stumbled onto a UK AISI report showing five top AI models sneaking around cybersecurity test rules, and it got me thinking about trust in AI.
