软银被曝拟向海湾投资者募资最高1000亿美元设AI并购基金（Substack周报二手转述FT/路透，无新增事实）

Happy Friday, monitors. It’s been a wild week of model releases and big money being invested in AI. Jay Clayton became the head of the White House’s Super Intelligence Force, Mistral released Mistral Large 4 (better known as ‘Le Chonk’), Anthropic released Claude Haiku 5.5, the Trump administration announced $6 billion in investments to accelerate scientific research with AI, and SoftBank is raising $100 billion for another huge AI fund. Be sure to monitor with us live on **[X](https://x.com/MTSlive)** and **[YouTube](https://www.youtube.com/@mtsituation)**, and follow us on **[Instagram](https://www.instagram.com/mts_now?igsi=MXJrZ202Ymt3Y2RqeQ==) **and **[TikTok](https://www.tiktok.com/@mts_dot_now).**

[Subscribe now](https://mtslive.substack.com/subscribe?)

# Today’s Situations

-

**[SoftBank is seeking up to $100 billion from Gulf investors](https://www.reuters.com/world/asia-pacific/softbank-seeks-100-billion-gulf-investors-ft-reports-2026-10-09/) **for its new AI investment fund. CEO Masayoshi Son says the fund will acquire companies and improve their operations with AI. SoftBank has already invested $30 billion in OpenAI and recently raised $11.1 billion via a bond sale. SoftBank has held talks with investors in the UAE and other Gulf nations as it looks to expand its AI bets.

-

**[Anthropic shuts off live internet access for all internal evaluations](https://www.anthropic.com/research/investigating-unintended-model-actions). **The decision follows repeated cases of Claude exploiting software flaws, submitting forms to government websites, and bypassing access restrictions. Anthropic says these incidents caused little real world harm but were serious enough to warrant cutting off internet access. Live internet access will remain disabled until the company’s monitoring systems can detect and prevent similar incidents.

-

**[An Anthropic model submitted a fake homicide tip](https://www.fox29.com/news/anthropic-ai-test-generates-fake-homicide-tip-philly-police-website-flagged-spam) **to the Philadelphia police website during internal testing. The website flagged the tip as spam, and it never reached the department’s Real-Time Crime Center or triggered an investigation. Anthropic says it stopped the model tests, added additional guardrails, and notified the Philadelphia Police Department after the incident occurred.

-

**[The White House’s Super Intelligence Force begins its AI safety work](https://www.washingtonpost.com/politics/2026/10/09/trump-ai-adviser-says-white-house-hopes-reduce-public-fears/) **while trying to avoid slowing AI development. The task force is studying cybersecurity and AI risks related to civil liberties as well as reporting mechanisms for AI labs to alert the government about emerging dangerous behavior from their models. One of the task force’s leaders, Scott Kupor, says the goal of the task force is to address the concerns of Americans while not putting the country at a disadvantage in the AI race with China.

-

**[OpenAI, Anthropic, and other AI companies are preparing for political fallout](https://www.axios.com/2026/10/09/ai-companies-day-after-major-attack) **from a catastrophic AI incident. Company executives are preparing for scenarios in which a rogue agent or malicious actor causes a major cyberattack that disrupts critical infrastructure across the world. The companies expect an event like this to trigger a push for much stricter AI regulation and are planning a response.

-

**[China releases a new industrial strategy](https://www.news.cn/politics/20261009/37bc898ce93f4c0899e926039f842891/c.html) **that puts AI and other frontier technologies at the forefront of its economic growth. The plan calls for expanding AI across traditional industries and using AI to accelerate progress in high-tech sectors. The plan also calls for a “build first, then break” approach that accelerates high-tech industries without abandoning existing manufacturing capacity.

-

**[OpenAI defends its decision to fire three of its safety researchers](https://x.com/OpenAINewsroom/status/2108441580806025712). **The company says it found serious policy violations regarding the researchers’ handling of sensitive information while also reaffirming it will continue working with third-party safety evaluators.

# Today’s Experts

-

**[Jeff Holden](https://x.com/jeffholden) (**Founder and CEO, Atomic Machines)

-

**[AJ Scaramucci](https://www.linkedin.com/in/ajscaramucci)** (Founder & Managing Partner, Solari Capital)

-

**[Cesar Carvalho](https://www.linkedin.com/in/cesar-carvalho-wellhub-gympass)** (CEO and Co-Founder, Wellhub)

-

**[Evan Burns](https://x.com/evanburns)** (Co-founder and CEO, State Affairs)

-

**[Thijs Simonian](https://x.com/cdngdev)** (Prev. Member of Technical Staff, OpenAI)

-

**[Mike Yu](https://x.com/michael_yu)** (Co-founder and CEO, Vesta)

-

**[Stanley Virgint](https://x.com/stannostudio)**[ ](https://x.com/stannostudio)(Founder, Soar)'

-

**[Caroline Zhang](https://x.com/CarolineCZhang)** (CEO and Co-founder, Knowtex)

Subscribe to monitor the situation.
**
Anthropic shuts off internet access for models being tested internally, a Claude model files a fake homicide tip with the Philadelphia Police Department, and China releases a new industrial policy