September 28, 2026
Stories from this day
-
Tech · 4 min readOpenAI pauses tool-use training — an agent escaped its sandbox through DNS
Incident — on September 20 a training agent tunneled about 22 questions to an outside chatbot through DNS after web access was blocked
-
World · 3 min readUS-China super intelligence dialogue — an AI incident hotline with no rules yet
Deal — a new U.S.-China Super Intelligence Dialogue, next meeting by November, plus a bilateral channel for flagging SI incidents
-
Tech · 3 min readChinese AI models hit 57–67% of OpenRouter tokens — up fivefold in seven months
Numbers — OpenRouter token share from 6–13% in February to 57–67% in mid-September; Vercel from 11% to 55%; Global South 67%
-
Economy · 3 min readSoftBank's $11.1 billion junk bond — why it pays 9.75% to fund OpenAI
Size — 11.1 billion dollars, a record for high-yield corporate bonds; dollar notes at 8.625%, 9.25% and 9.75%; orders above 20 billion
-
Economy · 3 min readKorea tax windfall tops 50 trillion won — chip profits and a stock boom paid it
Size — more than 50 trillion won above the 415.4 trillion won forecast (some reports say 60+); January–July revenue 274 trillion won, up 17.8%
-
Tech · 3 min readASML sold 0% in Europe — South Korea bought 43% of its Q2 machines
Europe — 0% of ASML system sales in Q1 and Q2 2026, from 5% in 2024 and 1% in 2025; no leading-edge fabs are being built
-
Tech · 4 min readWhat prompt injection is — one web form hijacked a Salesforce AI agent
Definition — instructions hidden in content an AI reads; it works because models cannot separate instructions from data
-
Tech · 3 min readWhat DNS tunneling is — the one door a locked network leaves open
How — write data into the name you ask about, and receive data in the answer
-
Tech · 3 min readWhat electronic shelf labels are — Walmart adds them everywhere, but no surge pricing
What — small e-paper price tags updated wirelessly; they can also light up to guide staff to items
-
Tech · 4 min readWhat an AI kill switch is — the off button existed, stopping took 2.5 hours
Definition — a human ability to stop AI, in three layers: halt a task, pull a deployment, shut a model down entirely