2026.10.01 AI news briefing — White House superintelligence pact and Google Gemini 4 Argon

At the White House, big tech executives signed a voluntary AI safety pact, and President Trump issued an executive order to use the term “superintelligence (SI)” instead of “AI.” On the same day, Google unveiled Gemini 4 Argon, claiming a return to the top of benchmarks, and in Korea, the government’s AI policy lineup convened together for the first time.

---

1. White House signs voluntary AI safety pact with big tech… executive order on the term “superintelligence (SI)”

• Source: CBS News | 2026.09.30
• Key points: President Trump and executives from more than 20 companies including OpenAI, Anthropic, Google, Meta, and Nvidia signed a voluntary AI safety evaluation pact at the White House. The core elements are establishing internal controls, external audits, and an oversight committee to review audit results. At the same event, President Trump also signed an executive order changing “AI” to “superintelligence (SI)” in federal documents.
• Why it matters: It lacks legal force, but following a string of agent runaway incidents, this is the largest industry-wide joint response to date and could set a reference point for future regulatory debates.

2. OpenAI unveils always-on personal agent “Dots”

• Source: TechCrunch | 2026.09.29
• Key points: The always-on agent “Dots,” built on GPT-6 Astra, has been announced. It runs independently on a dedicated cloud computer and browser, integrates with over 4,000 apps, and shares memory across ChatGPT, email, and Slack. It rolls out first to Pro and Business Premium, with the EEA, the UK, and Switzerland excluded.
• Why it matters: The arrival of an “always-on agent” as a formal product that tracks goals without human intervention is a key trend to watch alongside agent safety discussions.

3. Google launches frontier model “Gemini 4 Argon”

• Source: Google Blog | 2026.09.30
• Key points: This new model is specialized for coding, cyber defense, and workflow automation. It supports up to 1 million tokens of context, and according to Google’s published benchmarks, it outperforms GPT-6 Astra and Claude Opus 5.5 on several metrics. It is being released in limited form first to cybersecurity experts in the “Fairwind program,” priced at 2달러 per 1 million input tokens and 10달러 for outputs.
• Why it matters: While self-published benchmarks have limits, it signals Google’s renewed push into the frontier race after trailing OpenAI and Anthropic.

4. LinkAlpha begins supplying a “financial data AI marketplace” to Korean brokerages

• Source: 벤처스퀘어 | 2026.09.30
• Key points: New York–based financial AI startup LinkAlpha has begun supplying data to major Korean brokerages by processing 3.62 million+ disclosures and earnings materials into an AI-searchable, analyzable format. Each datapoint links to its source location, enabling verification of AI answers’ citations.
• Why it matters: As Korea’s financial sector deploys its own AI agents in production, verifiable source-grounded data quality is becoming a key adoption requirement.

5. Deputy PM Baekyung Hoon: “Not speed control, but speed responsibility”… first joint meeting of the government’s AI policy trio

• Source: 뉴스1 | 2026.09.30
• Key points: Minister of Science and ICT and Deputy Prime Minister 배경훈 chaired the 13th Ministers’ Meeting on Science and Technology, joined for the first time by 이해민, Senior Secretary to the President for AI Future Planning, and 하정우, Vice Chair of the National AI Strategy Committee. He said what’s needed is “not speed control but speed responsibility,” and a new “current issues meeting” was created for rapid response to urgent matters.
• Why it matters: It signals that Korea’s government AI policy lineup, vacant for four months, is now gearing up.

---

Today’s insights

• The US is effectively leaving the lead to industry through voluntary self-regulation and terminology redefinition, while Korea is consolidating government-led budget and coordination functions—a striking contrast on the same day.
• Both OpenAI’s and Google’s new agents/models are launching as “restricted releases with safety guardrails,” showing the industry is prioritizing control over speed after recent agent runaway incidents.